The roles of the Quasi-Biennial Oscillation and El Nino for entry stratospheric water vapour in observations and coupled chemistry-ocean CCMI and CMIP6 models
- 1Department of Physics, Ariel University, Ariel, Israel
- 2Eastern R&D center, Ariel, Israel
- 3The Fredy and Nadine Herrmann Institute of Earth Sciences, Hebrew University of Jerusalem, Jerusalem, Israel
- 4NOAA Chemical Sciences Laboratory, Boulder, CO, USA
- 5Cooperative Institute for Research in Environmental Sciences
- 1Department of Physics, Ariel University, Ariel, Israel
- 2Eastern R&D center, Ariel, Israel
- 3The Fredy and Nadine Herrmann Institute of Earth Sciences, Hebrew University of Jerusalem, Jerusalem, Israel
- 4NOAA Chemical Sciences Laboratory, Boulder, CO, USA
- 5Cooperative Institute for Research in Environmental Sciences
Abstract. The relative importance of two processes that help control the concentrations of stratospheric water vapor, the Quasi-Biennial Oscillation (QBO) and El Nino-Southern Oscillation (ENSO), are evaluated in observations and in comprehensive coupled ocean-atmosphere-chemistry models. The possibility of nonlinear interactions between these two is evaluated both using Multiple Linear Regression (MLR) and three additional advanced machine learning techniques. The QBO is found to be more important than ENSO, however nonlinear interactions are non-negligible, and even when ENSO, the QBO, and potential nonlinearities are included the fraction of entry water vapor variability explained is still substantially less than what is accounted for by cold point temperatures. While the advanced machine learning techniques perform better than an MLR in which nonlinearities are suppressed, adding nonlinear predictors to the MLR mostly closes the gap in performance with the advanced machine learning techniques. Comprehensive models suffer from too weak a connection between entry water and the QBO, however a notable improvement is found relative to previous generations of comprehensive models. Models with a stronger QBO in the lower stratosphere systematically simulate a more realistic connection with entry water.
Shlomi Ziskin Ziv et al.
Status: closed
-
RC1: 'Review', Anonymous Referee #1, 09 Feb 2022
The paper focusses on the factors affecting the interannual variability of stratospheric water vapor entry in the tropics in observations, CCMI and CMIP6 models. The authors contrast the use of a variety of techniques: multiple linear regression and 3 machine learning methods. Cold point temperatures are the main factor explaining the water vapor variability. They discuss the merits of the different techniques and the relative importance of the QBO and ENSO. They also find non-linear interactions to be important. The comprehensive models, whilst will suffering from a QBO that is not deep enough, have nonetheless improved. The paper is well written and provides an good description of machine learning techniques applied to a geophysical problem. The figures are also mostly clear.Â
Specific comments
(1) Make it clear earlier during the introduction that you are looking at interannnual variability and not the seasonal cycle.
(2) Some of the CCMI model have multiple ensembles. Do you average over all of them? If so, does this result in less variability and thus make it harder to compare to those runs with only 1 ensemble?
(3) In the figures, would it be possible to have the models with a nudged QBO labelled in bold text? It would make identifying them easier.
(4) On line 4, page 6, you mean ERA5/ERA5.1 I think?
(5) On page 6, line 11, "Note that the correlation of the BDC with the QBO is -0.66 (Figure 2), and hence including both in a single regression or ML model can lead to overfitting. " I disagree with this statement. Multicollinearity in your predictors causes a variety of problems but does not specifically cause overfitting. See page 283, Applied linear statistical models 5th edition by Neter et al. (2004). Your validation stage should show if overfitting is an issue.
(6) Page 10, line 15, the non-linear predictors are interesting but I struggle to relate them to physical processes. Could you give the reader a sense of what ENSO2Â might be?
(7) The values in Figure 6 are somewhat hard to read. Could you add a few labelled contour lines please?
(8) Figure 7 feels unnecessary since the same infomation can be conveyed with the text.
(9) In figure 9 (a to c), the text sugests that the solid black lines are observations (and they are not described in the caption) but where are there two parts and at different values? Label the models in 9(a).
Minor comments
Page 1, line 164, Emissions
Page2, line 5, through the its
Figure 1. Labels are a bit small and hard to read.
Figure 4. Are the units of the H20 anomalies correct?
Figure 5 and Figure 9. You use "std" and "std dev". Choose one to be consistent and also explain the abbreviation in the caption.
Figure 5(a) I am confused about the histogram. Is it normalised? If so, why are the values >1?
-
AC1: 'Reply on RC1', Shlomi Ziskin Ziv, 07 Apr 2022
The comment was uploaded in the form of a supplement: https://acp.copernicus.org/preprints/acp-2021-972/acp-2021-972-AC1-supplement.pdf
-
AC1: 'Reply on RC1', Shlomi Ziskin Ziv, 07 Apr 2022
-
RC2: 'Comment on acp-2021-972', Anonymous Referee #2, 11 Mar 2022
This paper discusses the importance of the nonlinear interaction between ESNO, QBO, and stratospheric water vapor, based on MLR and advanced machine learning techniques, and analyzes both observational data and chemistry-climate models. The authors conclude that QBO is more important than ENSO^2 than ENSO in predicting entry water vapor. The novel techniques and rigorous analysis of this paper will inspire the whole community, and I recommend this paper be accepted after a few revisions.
Â
General comments:
Â
- As the authors mentioned in line 5 and line 13 page 2, ENSO and QBO influences the stratospheric water vapor by influencing the tropical tropopause temperature. Later in Fig. 3, the authors compare the prediction of water vapor from merely tropical tropopause temperature, and from linear/nonlinear combination of ENSO and QBO. Since the ENSO and QBO directly influence tropical tropopause temperature and indirectly influences water vapor, before showing the relationship between ‘ENSO, QBO-stratospheric water vapor’, additional analysis of how well can linear/nonlinear combination of ENSO and QBO represents the tropical tropopause temperature will make the logic tighter.
- It is undoubted that considering the nonlinear process from ENSO and QBO can substantially increase the prediction of stratospheric water vapor, from the statistical analysis of this paper. However, more scientific arguments are needed when showing this result. For example, ENSO^2. The difference between ENSO and ENSO^2 are (1) ENSO^2 always amplifies extreme positive and negative ENSO states; (2) ENSO index has positive and negative values, but ENSO^2 only have magnitude, so extreme EN and LN will have similar ENSO^2 values. The authors explain (2) in section 3, but lack the necessary analysis of how (1) influences the predictions. Can you add another experiment of, say, absolute(ENSO)? It is possible that the behavior of abs(ENSO) is not as good as ENSO^2, since moderate events are not very important and ENSO^2 emphasizes the importance of extreme events so not necessary to add this experiment into the paper. Then I suggest that can add some more comments on page 13, lines 9-14 on how the two differences between ENSO and ENSO^2 improve the prediction. I also suggest including citations of why choosing ENSO^2 and ENSO*QBO not only in the introductions but also in result sections when discussing the improvement.
Â
Specific comments:
Â
- In figures showing the horizontal distributions, i.e., Fig.3, Fig.6, and Fig. 8, since ENSO is one of the most important topics of this paper, I suggest the base map should center at 180° instead of 0°, so the readers can compare the Western and Eastern Pacific more clearly.
- 10, please add panel numbers and titles.
- Page 1, line 15: please include more citations for ‘The amount of water vapor that enters the stratosphere is also important for stratospheric chemistry and specifically the severity of ozone depletion, for example, the citations on page 15, line 17.
- Page 4, line 21: ‘In total, more than 2500 year of model output are available’ I see no reason to calculate the total years because you are not putting all the model outputs together.
- Page 6, line 8: please introduce more about the radiosonde data, for example, is it monthly mean? Is the seasonal cycle included?
- Page 9, line 22: thanks for sharing, this is helpful to the community!
- Page 10, line 15: is the ‘busts’ problem in figure 4 still there in MLR2? 2010, 2015, and 2016 are all ENSO active years or right after so it is interesting to see whether adding ENSO^2 and QBO*ENSO can improve the performance or not.
- Page 17, line 15: ‘this results’ should be ‘this result’
-
AC2: 'Reply on RC2', Shlomi Ziskin Ziv, 07 Apr 2022
The comment was uploaded in the form of a supplement: https://acp.copernicus.org/preprints/acp-2021-972/acp-2021-972-AC2-supplement.pdf
Status: closed
-
RC1: 'Review', Anonymous Referee #1, 09 Feb 2022
The paper focusses on the factors affecting the interannual variability of stratospheric water vapor entry in the tropics in observations, CCMI and CMIP6 models. The authors contrast the use of a variety of techniques: multiple linear regression and 3 machine learning methods. Cold point temperatures are the main factor explaining the water vapor variability. They discuss the merits of the different techniques and the relative importance of the QBO and ENSO. They also find non-linear interactions to be important. The comprehensive models, whilst will suffering from a QBO that is not deep enough, have nonetheless improved. The paper is well written and provides an good description of machine learning techniques applied to a geophysical problem. The figures are also mostly clear.Â
Specific comments
(1) Make it clear earlier during the introduction that you are looking at interannnual variability and not the seasonal cycle.
(2) Some of the CCMI model have multiple ensembles. Do you average over all of them? If so, does this result in less variability and thus make it harder to compare to those runs with only 1 ensemble?
(3) In the figures, would it be possible to have the models with a nudged QBO labelled in bold text? It would make identifying them easier.
(4) On line 4, page 6, you mean ERA5/ERA5.1 I think?
(5) On page 6, line 11, "Note that the correlation of the BDC with the QBO is -0.66 (Figure 2), and hence including both in a single regression or ML model can lead to overfitting. " I disagree with this statement. Multicollinearity in your predictors causes a variety of problems but does not specifically cause overfitting. See page 283, Applied linear statistical models 5th edition by Neter et al. (2004). Your validation stage should show if overfitting is an issue.
(6) Page 10, line 15, the non-linear predictors are interesting but I struggle to relate them to physical processes. Could you give the reader a sense of what ENSO2Â might be?
(7) The values in Figure 6 are somewhat hard to read. Could you add a few labelled contour lines please?
(8) Figure 7 feels unnecessary since the same infomation can be conveyed with the text.
(9) In figure 9 (a to c), the text sugests that the solid black lines are observations (and they are not described in the caption) but where are there two parts and at different values? Label the models in 9(a).
Minor comments
Page 1, line 164, Emissions
Page2, line 5, through the its
Figure 1. Labels are a bit small and hard to read.
Figure 4. Are the units of the H20 anomalies correct?
Figure 5 and Figure 9. You use "std" and "std dev". Choose one to be consistent and also explain the abbreviation in the caption.
Figure 5(a) I am confused about the histogram. Is it normalised? If so, why are the values >1?
-
AC1: 'Reply on RC1', Shlomi Ziskin Ziv, 07 Apr 2022
The comment was uploaded in the form of a supplement: https://acp.copernicus.org/preprints/acp-2021-972/acp-2021-972-AC1-supplement.pdf
-
AC1: 'Reply on RC1', Shlomi Ziskin Ziv, 07 Apr 2022
-
RC2: 'Comment on acp-2021-972', Anonymous Referee #2, 11 Mar 2022
This paper discusses the importance of the nonlinear interaction between ESNO, QBO, and stratospheric water vapor, based on MLR and advanced machine learning techniques, and analyzes both observational data and chemistry-climate models. The authors conclude that QBO is more important than ENSO^2 than ENSO in predicting entry water vapor. The novel techniques and rigorous analysis of this paper will inspire the whole community, and I recommend this paper be accepted after a few revisions.
Â
General comments:
Â
- As the authors mentioned in line 5 and line 13 page 2, ENSO and QBO influences the stratospheric water vapor by influencing the tropical tropopause temperature. Later in Fig. 3, the authors compare the prediction of water vapor from merely tropical tropopause temperature, and from linear/nonlinear combination of ENSO and QBO. Since the ENSO and QBO directly influence tropical tropopause temperature and indirectly influences water vapor, before showing the relationship between ‘ENSO, QBO-stratospheric water vapor’, additional analysis of how well can linear/nonlinear combination of ENSO and QBO represents the tropical tropopause temperature will make the logic tighter.
- It is undoubted that considering the nonlinear process from ENSO and QBO can substantially increase the prediction of stratospheric water vapor, from the statistical analysis of this paper. However, more scientific arguments are needed when showing this result. For example, ENSO^2. The difference between ENSO and ENSO^2 are (1) ENSO^2 always amplifies extreme positive and negative ENSO states; (2) ENSO index has positive and negative values, but ENSO^2 only have magnitude, so extreme EN and LN will have similar ENSO^2 values. The authors explain (2) in section 3, but lack the necessary analysis of how (1) influences the predictions. Can you add another experiment of, say, absolute(ENSO)? It is possible that the behavior of abs(ENSO) is not as good as ENSO^2, since moderate events are not very important and ENSO^2 emphasizes the importance of extreme events so not necessary to add this experiment into the paper. Then I suggest that can add some more comments on page 13, lines 9-14 on how the two differences between ENSO and ENSO^2 improve the prediction. I also suggest including citations of why choosing ENSO^2 and ENSO*QBO not only in the introductions but also in result sections when discussing the improvement.
Â
Specific comments:
Â
- In figures showing the horizontal distributions, i.e., Fig.3, Fig.6, and Fig. 8, since ENSO is one of the most important topics of this paper, I suggest the base map should center at 180° instead of 0°, so the readers can compare the Western and Eastern Pacific more clearly.
- 10, please add panel numbers and titles.
- Page 1, line 15: please include more citations for ‘The amount of water vapor that enters the stratosphere is also important for stratospheric chemistry and specifically the severity of ozone depletion, for example, the citations on page 15, line 17.
- Page 4, line 21: ‘In total, more than 2500 year of model output are available’ I see no reason to calculate the total years because you are not putting all the model outputs together.
- Page 6, line 8: please introduce more about the radiosonde data, for example, is it monthly mean? Is the seasonal cycle included?
- Page 9, line 22: thanks for sharing, this is helpful to the community!
- Page 10, line 15: is the ‘busts’ problem in figure 4 still there in MLR2? 2010, 2015, and 2016 are all ENSO active years or right after so it is interesting to see whether adding ENSO^2 and QBO*ENSO can improve the performance or not.
- Page 17, line 15: ‘this results’ should be ‘this result’
-
AC2: 'Reply on RC2', Shlomi Ziskin Ziv, 07 Apr 2022
The comment was uploaded in the form of a supplement: https://acp.copernicus.org/preprints/acp-2021-972/acp-2021-972-AC2-supplement.pdf
Shlomi Ziskin Ziv et al.
Shlomi Ziskin Ziv et al.
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
393 | 104 | 19 | 516 | 10 | 10 |
- HTML: 393
- PDF: 104
- XML: 19
- Total: 516
- BibTeX: 10
- EndNote: 10
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1