the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 11 Jan 2024
Technical note: Challenges of detecting free tropospheric ozone trends in a sparsely sampled environment
Abstract. High quality long-term observational records are essential to ensure appropriate and reliable trend detection of tropospheric ozone. However, the necessity for maintaining high sampling frequency, in addition to continuity, is often under-appreciated. A common assumption is that so long as long-term records (e.g., span of a few decades) are available, (1) the estimated trends are accurate and precise, and (2) the impact of small-scale variability (e.g., weather) can be eliminated. In this study we show that the undercoverage bias (e.g., a type of sampling error resulting from statistical inference based on sparse or insufficient samples, such as once-per-week sampling frequency) can persistently reduce the trend accuracy of free tropospheric ozone, even if multi-decadal time series are considered. We use over 40 years of nighttime ozone observations measured at Mauna Loa, Hawaii (representative of the lower free troposphere) to make this demonstration, and quantify the bias in monthly means and trends under different sampling strategies. We also show that short-term meteorological variability remains a cause for an inflated long-term trend uncertainty. To improve the trend precision and accuracy due to sampling bias, two remedies are proposed: (1) a data variability attribution of colocated meteorological influence can efficiently reduce estimation uncertainty and moderately reduce the impact of sparse sampling; and (2) an adaptive sampling strategy based on anomaly detection enables us to greatly reduce the sampling bias, and produce more accurate trends using fewer samples compared to an intense regular sampling strategy.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(6848 KB)
-
Supplement
(6853 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(6848 KB) - Metadata XML
-
Supplement
(6853 KB) - BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Interesting statistical study on biases due to sparse sampling.', Anonymous Referee #1, 15 Feb 2024
This is an interesting study showing how sparse sampling may affect trends and monthly means of ozone in the troposphere. We all have to live with the limitations of our observing system. This study explicitly shows some of them, and also gives some ideas about possible improvements, by using additional parameters, or additional observations at suitable times. All of this is scientifically interesting and I commend the authors for a well written and well illustrated paper. Unfortunately, we also know that many observing systems are controlled by factors much different from the statistical properties of the observed atmospheric property, e.g. ozone. So while the study is very good and interesting, I wonder how much change, e.g. in ozone sounding frequency it could, or should, generate. Hopefully not a reduction.
Apart from this more general question, I don't really have any detailed comments. I think the authors have done a very good job, probably with a few iterations.
Citation: https://doi.org/10.5194/egusphere-2023-2739-RC1 -
CC1: 'Comment on egusphere-2023-2739', Raeesa Moolla, 28 Feb 2024
Overall, well written, in depth analysis, sound statistical methodological approach used. M
inor comments:
Page 3 - Line 19 - the word 'are' is missing (averages 'are' generally much smoother)
Page 11 - section 3.4 - Is it an option to increase frequency of reading and reduce spatial variability (i.e. less sondes, but more sampling days?
Page 12 - line 4 - Why 2-7 days and not 2-5 days? focus has been on 2-5 days mostly, so why analyse the cost-benefit of an additional 2 days ?
Citation: https://doi.org/10.5194/egusphere-2023-2739-CC1 -
RC2: 'Comment on egusphere-2023-2739', Anonymous Referee #2, 05 Mar 2024
This paper is a comprehensive sensitivity study of the impact of sampling rate on the accuracy and precision of inferred long-term trends. The paper’s syntax and structure are clear, although from time to time, the excessive amount of details makes it more difficult to follow.
The authors start with a high-elevation, multi-decadal surface ozone timeseries (Mauna Loa Observatory, 19.5N) known to have a high sampling frequency with very few gaps over several decades (the “perfect timeseries”), then design a number of scenarios in which only a fraction of all samples are used to compute the trends using classic multi-component LS and LAD fitting models (collocated dew point found to be the most critical proxy after deseasonalizing). The monthly mean bias, and trend bias and error are computed for all scenarios to assess which sub-sampling scenarios yield best agreement with the full-sampling results. The scientific approach is excellent and the authors try to address a very important and well-known aspect of atmospheric composition trend in general, which means that their findings are potentially applicable to a lot more studies beyond free tropospheric ozone.
If there must be one main criticism, it is only to say that the results are extremely detailed, too detailed at times, to a point that many figures could be simplified in order for the reader to extract the essential information. For example, I suggest that the "without meteorological adjustment" figures be removed after section 3.2. Another example, is the number of time periods, ranging from 1980-2021 to 2005-2021. I think showing results for only 2 or maximum 3 periods is just enough and will avoid overloading many figures.
The next main debatable point is the choice of ozone climatology for their so-called "adaptive sampling" strategy (D). Why choose 1980-1989 and not the entire period, or the period over which trends are computed? By choosing 1980-1989, it leaves room for a possible offset from the mean values that would be computed over the later periods, which potentially can skew the distribution of monthly mean bias on which strategy D is based upon.
Other comments are minor:
Figure 2 and related text (page 4):
There is no information on the measurement uncertainty for the MLO temperature and ozone surface instruments. Instruments yielding large measurement uncertainty are likely to produce similar inconsistency between calculated trends. The authors probably assume that the total measurement uncertainty of these instruments is small enough to be taken out of the equation. If so, please state it.
Figure 4, lines 17-29 (note on the sampling deviation vs. sampling bias):
This sentence is not clear. What does the expression “insufficient number of samples to infer a monthly mean value” mean exactly? Please re-phrase/clarify.
Figure 7:
Too much information between top and bottom row. I suggest top row shows only “with meteorological adjustment” trends, and only for period 1990-2021. This way reader has a much faster access to the actual information to be used from this figure (seasonality).
Figure 9:
Because panel (b) shows a coverage rate of 2.2 samples/week, I would suggest to show a 2-samples/week example in panel (a) instead of Sunday (1 day/week).
Figure S15:
It would be nice here to plot the trends from ozonesonde data itself, together with the MLO subsampled data. The consistency of these two independent datasets could be demonstrated, and would provide a direct justification to apply the methods discussed in this appear to the ozonesonde launch programs, among others. On the other end, if the two datasets do not show consistent results, this would trigger (a needed) discussion on the applicability of the method discussed in this paper, for example highlighting the possible impact of measurement uncertainty and long-term stability.
Page 14, lines 18-24: discussion on stratospheric intrusions:
Actually, stratospheric intrusions could (should?) be included among the “meteorological adjustments”. There is no scientific reasons to exclude samples underlying stratospheric intrusions and yet include samples underlying other dynamical/natural variability.
Citation: https://doi.org/10.5194/egusphere-2023-2739-RC2 - AC1: 'Response to reviewers: egusphere-2023-2739', Kai-Lan Chang, 30 Mar 2024
Interactive discussion
Status: closed
-
RC1: 'Interesting statistical study on biases due to sparse sampling.', Anonymous Referee #1, 15 Feb 2024
This is an interesting study showing how sparse sampling may affect trends and monthly means of ozone in the troposphere. We all have to live with the limitations of our observing system. This study explicitly shows some of them, and also gives some ideas about possible improvements, by using additional parameters, or additional observations at suitable times. All of this is scientifically interesting and I commend the authors for a well written and well illustrated paper. Unfortunately, we also know that many observing systems are controlled by factors much different from the statistical properties of the observed atmospheric property, e.g. ozone. So while the study is very good and interesting, I wonder how much change, e.g. in ozone sounding frequency it could, or should, generate. Hopefully not a reduction.
Apart from this more general question, I don't really have any detailed comments. I think the authors have done a very good job, probably with a few iterations.
Citation: https://doi.org/10.5194/egusphere-2023-2739-RC1 -
CC1: 'Comment on egusphere-2023-2739', Raeesa Moolla, 28 Feb 2024
Overall, well written, in depth analysis, sound statistical methodological approach used. M
inor comments:
Page 3 - Line 19 - the word 'are' is missing (averages 'are' generally much smoother)
Page 11 - section 3.4 - Is it an option to increase frequency of reading and reduce spatial variability (i.e. less sondes, but more sampling days?
Page 12 - line 4 - Why 2-7 days and not 2-5 days? focus has been on 2-5 days mostly, so why analyse the cost-benefit of an additional 2 days ?
Citation: https://doi.org/10.5194/egusphere-2023-2739-CC1 -
RC2: 'Comment on egusphere-2023-2739', Anonymous Referee #2, 05 Mar 2024
This paper is a comprehensive sensitivity study of the impact of sampling rate on the accuracy and precision of inferred long-term trends. The paper’s syntax and structure are clear, although from time to time, the excessive amount of details makes it more difficult to follow.
The authors start with a high-elevation, multi-decadal surface ozone timeseries (Mauna Loa Observatory, 19.5N) known to have a high sampling frequency with very few gaps over several decades (the “perfect timeseries”), then design a number of scenarios in which only a fraction of all samples are used to compute the trends using classic multi-component LS and LAD fitting models (collocated dew point found to be the most critical proxy after deseasonalizing). The monthly mean bias, and trend bias and error are computed for all scenarios to assess which sub-sampling scenarios yield best agreement with the full-sampling results. The scientific approach is excellent and the authors try to address a very important and well-known aspect of atmospheric composition trend in general, which means that their findings are potentially applicable to a lot more studies beyond free tropospheric ozone.
If there must be one main criticism, it is only to say that the results are extremely detailed, too detailed at times, to a point that many figures could be simplified in order for the reader to extract the essential information. For example, I suggest that the "without meteorological adjustment" figures be removed after section 3.2. Another example, is the number of time periods, ranging from 1980-2021 to 2005-2021. I think showing results for only 2 or maximum 3 periods is just enough and will avoid overloading many figures.
The next main debatable point is the choice of ozone climatology for their so-called "adaptive sampling" strategy (D). Why choose 1980-1989 and not the entire period, or the period over which trends are computed? By choosing 1980-1989, it leaves room for a possible offset from the mean values that would be computed over the later periods, which potentially can skew the distribution of monthly mean bias on which strategy D is based upon.
Other comments are minor:
Figure 2 and related text (page 4):
There is no information on the measurement uncertainty for the MLO temperature and ozone surface instruments. Instruments yielding large measurement uncertainty are likely to produce similar inconsistency between calculated trends. The authors probably assume that the total measurement uncertainty of these instruments is small enough to be taken out of the equation. If so, please state it.
Figure 4, lines 17-29 (note on the sampling deviation vs. sampling bias):
This sentence is not clear. What does the expression “insufficient number of samples to infer a monthly mean value” mean exactly? Please re-phrase/clarify.
Figure 7:
Too much information between top and bottom row. I suggest top row shows only “with meteorological adjustment” trends, and only for period 1990-2021. This way reader has a much faster access to the actual information to be used from this figure (seasonality).
Figure 9:
Because panel (b) shows a coverage rate of 2.2 samples/week, I would suggest to show a 2-samples/week example in panel (a) instead of Sunday (1 day/week).
Figure S15:
It would be nice here to plot the trends from ozonesonde data itself, together with the MLO subsampled data. The consistency of these two independent datasets could be demonstrated, and would provide a direct justification to apply the methods discussed in this appear to the ozonesonde launch programs, among others. On the other end, if the two datasets do not show consistent results, this would trigger (a needed) discussion on the applicability of the method discussed in this paper, for example highlighting the possible impact of measurement uncertainty and long-term stability.
Page 14, lines 18-24: discussion on stratospheric intrusions:
Actually, stratospheric intrusions could (should?) be included among the “meteorological adjustments”. There is no scientific reasons to exclude samples underlying stratospheric intrusions and yet include samples underlying other dynamical/natural variability.
Citation: https://doi.org/10.5194/egusphere-2023-2739-RC2 - AC1: 'Response to reviewers: egusphere-2023-2739', Kai-Lan Chang, 30 Mar 2024
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 393 | 112 | 21 | 526 | 37 | 11 | 11 |
- HTML: 393
- PDF: 112
- XML: 21
- Total: 526
- Supplement: 37
- BibTeX: 11
- EndNote: 11
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Owen R. Cooper
Audrey Gaudel
Irina Petropavlovskikh
Peter Effertz
Gary Morris
Brian C. McDonald
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(6848 KB) - Metadata XML
-
Supplement
(6853 KB) - BibTeX
- EndNote
- Final revised paper