|I (Andrew Sayer) reviewed the previous iteration of this manuscript, as well as the related Part I which was recently published in ACP. The authors have put a lot of effort into their revisions of both Part I and Part II which I feel improve the flow and content. This is an interesting and informative analysis and I think the authors have done a lot of good work; my comments below reflect the fact I want to help make the paper the best it can be, and there are a few points where the current text still has issues. I recommend another round of revisions and if the Editor would like would be happy to review the next version.|
There are three main areas where I think further changes are necessary:
1. Generally I think that the authors’ statements on policy attributions for changes in AOD still come across as too strong, since they aren’t directly examining the related emissions/meteorology patterns directly here (though they do refer to other studies for some of that information). From the authors’ response it seems that making strong attributive statements was not their intention so I wonder if some of this comes down to language differences on how we interpret the intensity of certain words. But for me, for example, the final paragraph of the abstract is a very strong attributive statement: “The long-term AOD variations presented here show that the changes in the emission regulations policy in China during 1995-2017 result in a gradual decrease of the AOD after 2011 with an average reduction of 30%-50% between 2011 and 2017. The effect is more visible in the highly populated and industrialized regions in SE China, as expected.” Similarly, the final paragraph of the paper: “Thus, in the current study the effect of the changes in the emission regulations policy in China is evident in the AOD decrease after 2011. The effect is more visible in the highly populated and industrialized regions in SE China.” I would suggest deleting these sentences entirely or at least replacing strong statements like “the variations … show that the changes in the emissions regulations policy … result in… “ and “the effect of the changes in the emission regulations policy in China is evident” as (for me as a reader) those words are making a direct explicit link, which could be true, but is not supported by the analysis in the study. It would absolutely need additional analysis including emissions estimates, satellite/ground data on trace gas precursors for aerosols like SO2/NOx, meteorology etc. Yes I know that other studies have looked at these in isolation but if you want to make a statement about attribution you need to show it all together. That could be a follow-up study. I don’t see why there is a need to make a strong statement about attribution in this study anyway?
2. One thread of my previous review was to better understand the uncertainties in the trend calculation resulting from combining the time series from the two (quite different) retrievals. The authors have not been able to quantify the additional uncertainty which this adds, noting (correctly) that this is not a simple task to do so. I also suggested they adopt the methodology of Wilks to assess/remove false positives in the significance calculations due to multiply hypothesis testing, which they chose not to do. In light of both of the above, I think that discussion about statistical significance (here I think it is based on trend/tendency magnitude being more than twice the estimated uncertainty on it, which is saying that the chance of this happening if there is no trend is 5% or less) should be removed from the paper. The p<0.05 approach as an arbitrary threshold for whether something is statistically significant is so well-used that it has become a gatekeeper for whether or not a given result is talked about (and other research communities such as psychology/medicine are currently undergoing something of a paradigm change in terms of dealing with this), and in this case we know (and the authors have stated in their response) that there are components of the trend uncertainty that they do not include because they cannot model them well. In that case we know that the total uncertainties are likely to be underestimated by unknown magnitude, and as a result estimates of the statistical significance of the result are overestimated by an unknown magnitude. Because of this I think that it is better to present not trend estimates and their significance but rather trend estimates and their best estimate of the uncertainty, and to acknowledge that this uncertainty is a lower bound due to these issues. This would obviously require changes to the manuscript throughout (to text, figures, and tables; maybe for some figures panels showing uncertainty could be added, or filters for maps could be based on absolute uncertainty in the trend rather than an estimated significance level), but seems to be a fairer presentation which avoids a reader potentially interpreting a result as stronger (as in, more likely real than due to chance) than it is. I do not see an advantage in including the significance values when we know that they are biased estimates of significance and know that they may end up misleading a reader who is not statistically-minded or does not go through thoroughly.
3. As the authors know I am strongly against least-squares linear regression for AOD validation scatter plots, because although it is a common technique it is statistically invalid here, as the data violate most/all of the requirements for the technique to be applicable (see e.g. http://people.duke.edu/~rnau/regintro.htm or statistics textbooks). The slope/intercept coefficents and estimated uncertainties you get up are known to be wrong and systematically (not randomly) biased in such cases. I don’t see any scientific justification for including something we know to be wrong in the paper. I also don’t think they are a useful interpretive aid for e.g. Figure 5, since the data volume is fairly low, and the binned analysis (circles with error bars) here convey related information (i.e., magnitude and sign of the bias in low-AOD and high-AOD conditions) in a much better way. Please just delete the regression slopes/intercepts and discussion from the paper.
I also have a number of more minor issues:
P2L21: I don’t think that “total suspended particle matter (TSP)” is needed here – it would be better just to write “particulate matter” and the grammar isn’t quite right written this way (“Coal smoke mainly contains total suspended particle matter” doesn’t make sense while “Coal smoke mainly contains particulate matter” does).
Figure 2: based on the text, the labels for the Taklamakan Desert and Tibetan Plateau (8 and 9) are inconsistent. (I think this was also the case in the original submission but I did not spot the error then.)
P8L14: I think that there are some missing numbers about fine-dominated error here (looks like only the coarse condition results are shown), based on the later text on P9 L30-31. These two sections are also repetitive; I’d just delete this sentence on P8L14 as I think the information fits better on P9.
Figures 7, 8 (as well as general comment on the text throughout): in addition to my significance comment above, these two figures are somewhat repetitive in that it’s the same thing just expressed on an absolute scale in Figure 7 and a relative scale in Figure 8. I am not sure that Figure 8 adds much, and suggest deleting it. A small (and scientifically unimportant) change over a fairly clean background can appear prominent as a small change to a small number can appear big, while a more modest increase/decrease over a polluted area (which might represent a real policy success or failure) in contrast fades from prominence as it becomes a smaller relative change. In other words, getting 100% richer is quite different in real terms if you only had $1 to start with compared with if you were a billionaire to start with. Basically I’m saying that Figure 8 could distract from where the interesting results are happening. In general I am in favour of absolute rather than relative AOD changes throughout for that reason (and I think the other reviewer had expressed comments against writing things in relative terms). Removing the stuff about relative changes would also make text and figures more readable by making the sentences more concise and removing visual clutter. I don’t see an advantage as a reader to including both.