|This manuscript reports on measurements of aerosol sulfur in aerosol samples collected from the IAGOS-CARIBIC platform over a 16 year period. Analysis focuses on a new regression technique that the authors suggest can be used to infer both the gradient of sulfur concentration and the integrated burden in the lowermost stratosphere (LMS), starting at the dynamical tropopause and extending to 3 km above it. It is also suggested that this analysis provides an estimate of the relative contribution of stratospheric sulfate mixed downward and tropospheric sources on the sulfur concentration in the upper troposphere (UT), and how these contributions vary seasonally. Compared to the original version, changes in the data processing now produce more pronounced seasonality in the estimated contribution of S derived from tropospheric sources to the S burden in the UT. In particular, the analysis now suggests a quite strong peak which authors attempt to link to downward transport from the ATAL.|
While the approach has changed in response to comments on first version, and some sections describing the data analysis are now more clear, I still find myself not very convinced. Step 1, doing forced linear regression (FLR) of S versus distance above the tropopause is relatively straight forward, except for one key point. The data “were grouped with respect to concentration levels of the different years, resulting in 4 to 5 groups of data for each season” (lines 179-180). The discussion about this step seems to justify at least 2 groups (background and influenced by volcanoes), and perhaps the volcanic group could be (was?) further divided into strong and weak influence. It seems that the authors recognized that there had to be more than 2 groups in each season to do step 2 regressions, but how these were created is not explained. Some objective method of creating these groups, perhaps based on mean, median, max measured S concentrations, or maybe based on time since eruptions were known to have perturbed the LMS burden (based on previous work by this group using largely the same data set) would seem essential, especially for anyone attempting to replicate the analysis or apply the technique to some future similar data set. As it reads now, it almost seems that the approach was to sort by year based on timing of volcanos, but then move years (or individual samples?) that didn’t work into another group until things looked better
The next step in the analysis is now basically clear conceptually: for each season the intercept and slope from 4 or 5 FLR in step 1 were regressed against each other, using weighted regression with the weighting based on estimated uncertainty in the FLR fits. The explanation of this weighting is not completely clear to me, but seems to generate uncertainty only in the FLR intercept (termed offset) when transferred to Fig 3 which shows results of the weighted regressions by season. Surely, the slopes from step 1 were also uncertain. It would seem that uncertainty in both slope and intercept from the FLR would propogate into uncertainty in the estimated slope and intercept from the weighted regression. Based on the small number of data points and the magnitude of the error bars in all 12 panels of Fig 3, I find the small error bars in Fig 5 very surprising (seems that error bars in 5 a should come directly from uncertainty in the slope from fits in Fig 3, while those in 5 b come from uncertainty in the intercepts in Fig 3). In particular, the authors need to carefully consider how to propagate uncertainty from both initial measurement, uncertainty in Z, and uncertainty in slope and intercept of the FLR to the intercept values for the weighted regressions in Fig 3 and clearly explain how they do this, since they assert that the derived values represent the concentration of S in the UT that is derived from tropospheric sources and use these numbers for most of the remaining analyses in the paper. Lines 244 – 249 address how they attempted to account for the uncertainty from FLR, but not in very clear manner.
If the seasonality shown in Figures 5 and 7 is real (signal truly larger than uncertainty) that is an interesting finding. Going to CALIOP data to try to explain the fall peak in tropospheric influence on S in UT is also a good idea, but I am not sure how much support the progressive descent of the ATAL in the top row of Fig 8 provides to the athors interpretation of CARIBIC UT/LS S. How much does the ATAL spread zonally over just a few months? Would the upper level anticyclone associated with the Asian monsoon not keep the enhanced aerosol more or less over Tibet? Did many (any) of the CARIBIC flight pass through or near the 60-120E longitude band shown in Fig 8?
In summary, I am still not convinced that the data analysis is robust enough to believe that from now forward every CARIBIC S measurement can be converted into an estimate of the S column 3 km above the tropopause (Fig 6), or that the source of S in the UT changes from almost completely stratospheric in FM to about 90% tropospheric in ON (Fig. 7). Authors need to use and describe an objective means to bin their data, and more clearly discuss how confident they are about the uncertainty in their derived estimates of “particulate sulfur concentration of tropospheric origin” (also known as the intercepts of the weighted regressions).
Minor editorial comments
Line 82 what is meant by “crust and particle-bound water”?
Lines 147-149 reword this to make it clear what is meant by “not temperature decrease in relation to surface of the earth or the degree of cloud processing” and how either of these is related to distance below the PV tropopause.
Lines 178-194 This section is where the data binning and how 4 or 5 groups were objectively defined needs to be explained.
Line 223 and equation 2, seems that year (y) should be group (g). If you really did the regression for 12 seasons and 16 years the number would be >> 52
Line 284 Fig 4--→Fig 4a
Line 289 Fig 4-→Fig 4b
Line 415-417 This sentence is describing the seasonality of “particulate sulfur concentration of tropospheric origin” not the concentration of S in UT. Figure 7 shows nearly no seasonal change in UT S concentration in background year, and a spring peak in the volcanic year.