the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 29 Apr 2024
Contrail altitude estimation using GOES-16 ABI data and deep learning
Abstract. The climate impact of persistent aircraft contrails is currently estimated to be comparable to that due to aviation-emitted CO2. A potential near-term and low-cost mitigation option is contrail avoidance, which involves re-routing aircraft around ice supersaturated regions, preventing the formation of persistent contrails. Current forecasting methods for these regions of ice supersaturation have been found to be inaccurate when compared to in situ measurements. Further assessment and improvements of the quality of these predictions can be realized by comparison with observations of persistent contrails, such as those found in satellite imagery. In order to further enable comparison between these observations and contrail predictions, we develop a deep learning algorithm to estimate contrail altitudes based on GOES-16 ABI infrared imagery. This algorithm is trained using a dataset of 3267 contrails found within CALIOP LIDAR data and achieves a root mean square error of 570 m. The altitude estimation algorithm outputs probability distributions for the contrail top altitude in order to represent predictive uncertainty. The 95 % confidence intervals constructed using these distributions, which are shown to contain approximately 95 % of the contrail data points, are found to be 2.2 km thick on average. These intervals are found to be 34.1 % smaller than the 95 % confidence intervals constructed using flight altitude information alone, which are 3.3 km thick on average. Furthermore, we show that the contrail altitude estimates are consistent in time and, in combination with contrail detections, can be used to observe the persistence and three-dimensional evolution of contrail forming regions from satellite images alone.
-
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
(7151 KB)
-
Supplement
(6511 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(7151 KB) - Metadata XML
-
Supplement
(6511 KB) - BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2024-961', Ziming Wang, 26 May 2024
This paper presents the first remote-sensing-based contrail altitude estimation algorithm. Both the image-level model and the cirrus pixel-by-pixel model are developed and compared, with an evaluation of predictive uncertainty and an assessment of the method's accuracy using individual test data and independent flight data. This study offers valuable insights for further assessing the climate impact of contrail cirrus. The paper is well-organized and well-written. I urge its publication in AMT, with some minor comments provided for the authors' consideration.
Specific comments:
Line 40: Please provide the physical explanations for why the infrared channels are used for estimating cloud top altitude.
Figure 7: The plot shows a trend where the CNN generally overestimates contrail altitude compared to the true values from CALIPSO. Are there any potential ideas for this?
Figure 10: The plot here seems to support my impression from Figure 7 that the contrail altitude can be slightly overestimated. During data collocation, you carefully considered the advection of aircraft data due to horizontal wind. Then, contrail ice crystals can sediment, which should theoretically reduce the altitude rather than increase it when compared to the flight data. Are there any reasons behind this discrepancy?
Conclusion: The RMSE is used as the metric to indicate the accuracy of the algorithm, as emphasized in the abstract. Since the developed contrail altitude retrieval method is the next step due to the biased prediction of ice supersaturation vertical extension in contrail avoidance, would it be better to also show the simple mean bias error or mean absolute error for estimating the contrail altitude?
Technical corrections:
Caption of Figure 1: "Zulu" time is equivalent to "UTC" time. However, I'm not sure if it is widely used in this research field. This applies to the entire text to be consistent with the figure.
L90: “a 50km distance of the ground-track of CALIPSO.” I assume it should refer to the supplement S1.
L132: “FlightAware (for times in 2023)”. Eventually it appears not to have been used because the focus was on the years 2018-2022.
L221: “ISS” instead of “ISSRs”.
L273: tends to be over-confident for probabilities between 0.5 and 0.9, as well as between 0.1 and 0.2.
Overall, the excellent work presented in this article is acknowledged.
Citation: https://doi.org/10.5194/egusphere-2024-961-RC1 - AC1: 'Reply on RC1', Vincent R. Meijer, 09 Jul 2024
-
RC2: 'Comment on egusphere-2024-961', Anonymous Referee #2, 03 Jun 2024
Summary:
The authors use two deep learning techniques to estimate contrail cloud top height in GOES infrared imagery trained with CALIOP lidar data. Over 3000 contrails over five-year period are collocated with the lidar data to allow for the building and testing of the deep learning methods. The more successful contrail height estimate method was developed from a convolutional neural network (CNN), estimating contrail height with a root mean square error of 570 m in test data. An analysis of the CNN method results show that the predictive probability of the CNN method is generally well calibrated and has a smaller 95% confidence interval than the confidence intervals derived from flight altitude data alone. The authors also processed a 24-hour period of GOES data to show the spatial and temporal distribution of the contrail height estimates.
General comments:
The overall quality of the manuscript is good.The authors explain the methodology and results of the research concisely, and reach logical and consistent conclusions. Although the topic of the manuscript is contrail height estimation, the authors include discussion of thin cirrus height estimation that is unnecessary (considering the title of the manuscript) and confusing (especially subsection 3.2). Unless the authors can show why the cirrus altitude estimation is integral to the research presented in the main manuscript, I suggest that discussion about cirrus altitude estimation be removed from the paper. The authors make multiple references to the Supplementary Materials, so much so that it is nearly impossible to understand the manuscript without also reading those pages. As a result, the manuscript is incomplete and might not stand alone. The reader should not have to rely on the Supplementary Materials to read the principal paper. Finally, the paper lacks references to multiple concepts that should be explained in the paper (not just the Supplementary Materials). I can find no references for the various height conversions (between geometric, geopotential, and pressure altitudes), the advection of contrails, or the parallax correction used in the main paper. Add these references to the manuscript.
Specific comments:
Lines 100-104: Could not the width of contrail 2 also be the result of the geometry of the contrail relative to the CALIPSO ground track? Most of the other contrails are nearly perpendicular to the ground track, while the angle between contrail 2 and the satellite track is much more acute?
Lines 109-117: This paragraph is unclear.. The authors state “The collocation process is nearly identical at that for contrails, except that the contrail detection masks are no longer involved.” I can’t find any mention of cloud masks up to this point.
Line 156: What are “normal” operations? Even after reading the paper, it is not clear to me what that means.
Lines 172-173: ‘…all inputs to the neural networks are “observational”…’ What is “observational” in this context? Why not say instead that the inputs are derived from the satellite radiances alone with no additional (NWP data) used?
Figure 2 and elsewhere: It is not always clear which altitude the authors are using. In Figure 2, for example, what type of altitudes are plotted here? Geometric? Geopotential? Pressure altitudes?
Lines 194-195: What is the “thickness” of the altitude distribution? It appears in Figure 2 that the variance of the altitude distribution increases as the latitude increases, contrary to the text.
Figure 3: Why are the contrail tops generally so much higher than the ISS & SAC regions (except for summer)? It is apparent from the paper that the three profiles (Flight, ISS & SAC, Contrail top) represent entirely different times, locations, and number of observations. It would be better to make this distinction much more clear to the reader, otherwise they may be confused by this figure.
Section 3.2: The authors refer much more to the Supplementary Materials here than the manuscript itself. Many of the values stated in the text don’t match any of the values presented in Figure 4. This is very confusing! As stated earlier, I suggest the authors remove any discussion of cirrus altitude from the paper. It is superfluous and not presented well.
Figure 5: Green line, blue line, black line. Which models do they represent? A legend would make this figure much easier to understand.
Line 311: Why is “simulate” in quotation marks? It appears to be a simulation (i.e, it imitates the appearance of) in the true sense of the word. The parallax correction is actually making the alignment of the flight tracks match better with the detected contrails.
Figure 7: Unless they looked that the Supplementary Materials, the reader would not know what “% of distance flown in 2 hours before” would mean. Some description of how this quantity was obtained must be included in the manuscript, not just the Supplementary Materials.
Citation: https://doi.org/10.5194/egusphere-2024-961-RC2 - AC2: 'Reply on RC2', Vincent R. Meijer, 09 Jul 2024
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2024-961', Ziming Wang, 26 May 2024
This paper presents the first remote-sensing-based contrail altitude estimation algorithm. Both the image-level model and the cirrus pixel-by-pixel model are developed and compared, with an evaluation of predictive uncertainty and an assessment of the method's accuracy using individual test data and independent flight data. This study offers valuable insights for further assessing the climate impact of contrail cirrus. The paper is well-organized and well-written. I urge its publication in AMT, with some minor comments provided for the authors' consideration.
Specific comments:
Line 40: Please provide the physical explanations for why the infrared channels are used for estimating cloud top altitude.
Figure 7: The plot shows a trend where the CNN generally overestimates contrail altitude compared to the true values from CALIPSO. Are there any potential ideas for this?
Figure 10: The plot here seems to support my impression from Figure 7 that the contrail altitude can be slightly overestimated. During data collocation, you carefully considered the advection of aircraft data due to horizontal wind. Then, contrail ice crystals can sediment, which should theoretically reduce the altitude rather than increase it when compared to the flight data. Are there any reasons behind this discrepancy?
Conclusion: The RMSE is used as the metric to indicate the accuracy of the algorithm, as emphasized in the abstract. Since the developed contrail altitude retrieval method is the next step due to the biased prediction of ice supersaturation vertical extension in contrail avoidance, would it be better to also show the simple mean bias error or mean absolute error for estimating the contrail altitude?
Technical corrections:
Caption of Figure 1: "Zulu" time is equivalent to "UTC" time. However, I'm not sure if it is widely used in this research field. This applies to the entire text to be consistent with the figure.
L90: “a 50km distance of the ground-track of CALIPSO.” I assume it should refer to the supplement S1.
L132: “FlightAware (for times in 2023)”. Eventually it appears not to have been used because the focus was on the years 2018-2022.
L221: “ISS” instead of “ISSRs”.
L273: tends to be over-confident for probabilities between 0.5 and 0.9, as well as between 0.1 and 0.2.
Overall, the excellent work presented in this article is acknowledged.
Citation: https://doi.org/10.5194/egusphere-2024-961-RC1 - AC1: 'Reply on RC1', Vincent R. Meijer, 09 Jul 2024
-
RC2: 'Comment on egusphere-2024-961', Anonymous Referee #2, 03 Jun 2024
Summary:
The authors use two deep learning techniques to estimate contrail cloud top height in GOES infrared imagery trained with CALIOP lidar data. Over 3000 contrails over five-year period are collocated with the lidar data to allow for the building and testing of the deep learning methods. The more successful contrail height estimate method was developed from a convolutional neural network (CNN), estimating contrail height with a root mean square error of 570 m in test data. An analysis of the CNN method results show that the predictive probability of the CNN method is generally well calibrated and has a smaller 95% confidence interval than the confidence intervals derived from flight altitude data alone. The authors also processed a 24-hour period of GOES data to show the spatial and temporal distribution of the contrail height estimates.
General comments:
The overall quality of the manuscript is good.The authors explain the methodology and results of the research concisely, and reach logical and consistent conclusions. Although the topic of the manuscript is contrail height estimation, the authors include discussion of thin cirrus height estimation that is unnecessary (considering the title of the manuscript) and confusing (especially subsection 3.2). Unless the authors can show why the cirrus altitude estimation is integral to the research presented in the main manuscript, I suggest that discussion about cirrus altitude estimation be removed from the paper. The authors make multiple references to the Supplementary Materials, so much so that it is nearly impossible to understand the manuscript without also reading those pages. As a result, the manuscript is incomplete and might not stand alone. The reader should not have to rely on the Supplementary Materials to read the principal paper. Finally, the paper lacks references to multiple concepts that should be explained in the paper (not just the Supplementary Materials). I can find no references for the various height conversions (between geometric, geopotential, and pressure altitudes), the advection of contrails, or the parallax correction used in the main paper. Add these references to the manuscript.
Specific comments:
Lines 100-104: Could not the width of contrail 2 also be the result of the geometry of the contrail relative to the CALIPSO ground track? Most of the other contrails are nearly perpendicular to the ground track, while the angle between contrail 2 and the satellite track is much more acute?
Lines 109-117: This paragraph is unclear.. The authors state “The collocation process is nearly identical at that for contrails, except that the contrail detection masks are no longer involved.” I can’t find any mention of cloud masks up to this point.
Line 156: What are “normal” operations? Even after reading the paper, it is not clear to me what that means.
Lines 172-173: ‘…all inputs to the neural networks are “observational”…’ What is “observational” in this context? Why not say instead that the inputs are derived from the satellite radiances alone with no additional (NWP data) used?
Figure 2 and elsewhere: It is not always clear which altitude the authors are using. In Figure 2, for example, what type of altitudes are plotted here? Geometric? Geopotential? Pressure altitudes?
Lines 194-195: What is the “thickness” of the altitude distribution? It appears in Figure 2 that the variance of the altitude distribution increases as the latitude increases, contrary to the text.
Figure 3: Why are the contrail tops generally so much higher than the ISS & SAC regions (except for summer)? It is apparent from the paper that the three profiles (Flight, ISS & SAC, Contrail top) represent entirely different times, locations, and number of observations. It would be better to make this distinction much more clear to the reader, otherwise they may be confused by this figure.
Section 3.2: The authors refer much more to the Supplementary Materials here than the manuscript itself. Many of the values stated in the text don’t match any of the values presented in Figure 4. This is very confusing! As stated earlier, I suggest the authors remove any discussion of cirrus altitude from the paper. It is superfluous and not presented well.
Figure 5: Green line, blue line, black line. Which models do they represent? A legend would make this figure much easier to understand.
Line 311: Why is “simulate” in quotation marks? It appears to be a simulation (i.e, it imitates the appearance of) in the true sense of the word. The parallax correction is actually making the alignment of the flight tracks match better with the detected contrails.
Figure 7: Unless they looked that the Supplementary Materials, the reader would not know what “% of distance flown in 2 hours before” would mean. Some description of how this quantity was obtained must be included in the manuscript, not just the Supplementary Materials.
Citation: https://doi.org/10.5194/egusphere-2024-961-RC2 - AC2: 'Reply on RC2', Vincent R. Meijer, 09 Jul 2024
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 413 | 218 | 25 | 656 | 56 | 16 | 15 |
- HTML: 413
- PDF: 218
- XML: 25
- Total: 656
- Supplement: 56
- BibTeX: 16
- EndNote: 15
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Vincent R. Meijer
Sebastian D. Eastham
Ian A. Waitz
Steven R.H. Barrett
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(7151 KB) - Metadata XML
-
Supplement
(6511 KB) - BibTeX
- EndNote
- Final revised paper