the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 01 Nov 2022
TIMBER v0.1: a conceptual framework for emulating temperature responses to tree cover change
Abstract. Society is set to experience significant land cover changes in order to achieve the temperature goals agreed upon under the Paris Agreement. Such changes carry both global implications, pertaining to the biogeochemical effects of land cover change and thus the global carbon budget, and regional/local implications, pertaining to the biogeophysical effects arising within the immediate area of land cover change. Biogeophysical effects of land cover change are of high relevance to national policy- and decision- makers and their accountance is essential towards effective deployment of land cover practices that optimises between global and regional impacts. To this end, ESM outputs that isolate the biogeophysical responses of climate to land cover changes are key in informing impact assessments and supporting scenario development exercises. Generating multiple such ESM outputs, in a manner that allows comprehensive exploration of all plausible land cover scenarios however, is computationally untenable. This study proposes a framework to agilely explore the local biogeophysical responses of climate under different land cover scenarios by means of a computationally inexpensive emulator. The emulator is novel in that it solely represents the land cover forced, biogeophysical responses of climate, and can be used as either a standalone device or supplementary to existing climate model emulators that represent greenhouse gas (GHG)- or Global Mean Temperature (GMT)- forced climate responses. We start off by modelling local minimum, mean and maximum surface temperature responses to tree cover changes by means of a month- and Earth System Model (ESM)- specific Generalised Additive Model (GAM) trained over the whole globe. 2-m air temperature responses are then diagnosed from the modelled minimum and maximum surface temperature responses using observationally derived relationships. Such a two-step procedure accounts for the different physical representations of surface temperature responses to tree cover changes under different ESMs, whilst respecting a definition of 2-m air temperature that is more consistent across ESMs and with observational datasets. In exploring new tree cover change scenarios, we employ a parametric bootstrap sampling method to generate multiple possible temperature responses, such that the uncertainty within the GAM's derived shape of the response is also quantified. The output of the final emulator is demonstrated for the SSP 1-2.6 and 3-7.0 scenarios. Relevant temperature responses are identified as those displaying a clear signal in relation to the surrounding uncertainty in shape of derived response, calculated as the "signal-to-noise" ratio between the sample set mean and sample set variability. The emulator framework developed in this study thus provides a first step towards bridging the information-gap surrounding biogeophysical implications of land cover changes, allowing for smarter land-use decision making.
-
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
(6727 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(6727 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2022-1024', Anonymous Referee #1, 05 Dec 2022
General comments:
This manuscript describes an emulation of earth system model output, specific to the biogeophysical responses to forest cover change. Afforestation, reforestation, and reference scenarios are used to assess the model's representation of change in air temperature at and 2 metres above the land surface. The emulator appears to perform well against MPI-ESM, but substantial errors accumulate for extreme afforestation and deforestation scenarios in the other models and against observations.
The research topic is an important and relevant one, as the computational complexity of ESMs increases, as does the importance of assessing and planning land-based climate mitigation responses. The study is well-designed and clearly argued and the manuscript is well-structured and thorough. Overall I would recommend its publication with minor revisions.
Specific comments:
What about anything other than surface and 2m air temperature? At least acknowledge its limitations as a metric for all "biophysical" climate impacts (e.g. that it conflates albedo- and latent heat-related surface temperature changes despite their quite different effects on the atmospheric energy balance).
There is no mention of tree species/PFT and how the variation in tree types around the world influences biophysical properties in different biomes. Since tree PFT likely accounts for a major part of the variance of biophysical properties in the ESMs, I would have liked to know how this was translated to the emulator.
The figure captions should fully define all acronyms and variable names. It should not be necessary to read the text in detail to gain a basic understanding of what the figures show.
Technical corrections:
L5: usage: suggest "accounting for them" to replace "their accountance"
L9 and L413: usage: suggest removing "agilely", which does not feel like a natural word, and its meaning is communicated later in the sentence already (with "by means of a computationally inexpensive manner"
L57: usage: suggest "hereafter" to replace "hereon"
L59: typo: "biobphysical"Citation: https://doi.org/10.5194/egusphere-2022-1024-RC1 -
RC2: 'Comment on egusphere-2022-1024', Anonymous Referee #2, 22 Dec 2022
Review of egusphere-2022-1024:
TIMBER v0.1: a conceptual framework for emulating temperature responses to tree cover change
Summary
The authors present a new approach for estimating the local surface temperature responses to changes in tree cover. They first fit and validate multiple GAMs to model outputs and observations (separately). They also generate additional GAMs to represent parametric uncertainty in these models. They asses the validity of using an existing observational relationship to determine 2-m surface temperature from ESM estimated min and max surface temperatures. They show reliable GAM results and state that the existing relationship is good enough for this study. These tools are then applied to two SSP scenarios of tree cover change to demonstrate the utility of this approach for estimating temperature change due to change in tree cover.
Overall review
I appreciate this effort and see this as a good contribution toward understanding the potential effects of land cover change on climate. The authors employ a clever mix of statistical techniques in a thorough framework that estimates surface temperature change due to tree cover change. My main concerns are with the paper organization and incomplete description of methods. Overall the methods are sound (as far as I can tell), although one aspect in particular could be improved, and comparison with an alternative method is warranted. I recommend addressing the following three main issues. See below for detailed comments.
1) The methods are incomplete. Some methods are out of place while others are missing or unclear.
2) I am not convinced that the Hooker relationship as used is adequate for the final estimation of 2m T. For the time being, it shows the potential of this overall approach, but it seems to have a rather large spread of error for operational use. The authors should consider taking up their suggestion and doing a model-specific fit of Tmin/max to 2m T, and comparing this to the current results.
3) The scenario results should be compared with other estimates of temperature response to land cover change. There are comparable examples of this using these three ESMs. While such examples use a simulation differencing method that encapsulates much more than the method in this study, One would expect some similarities given the dominant role of tree cover change in surface temperature change. Maybe the comparison is between the TSmean change estimated by the GAMs for each scenario, rather than the 2m T, and the mean surface temperature change estimated by model differencing.
Specific suggestions and comments
Abstract
line 1:
“Society is set to experience…” is a bit overstated here. Evidence shows that the world is not even close to meeting the Paris Agreement. I suggest revising to something like “Land cover change has been proposed to play a significant role in achieving the temperature goals…”
line 6:
expand ESM
1. Introduction
lines 32-33:
you probably want to include the boysen 2020 paper in this citation also.
2. Training datasets
line 75-77:
Is there just one REF scenario? Then “The REF scenario spans 150 years…”
What do you mean by “full expansion?” Is 100% of the surface covered by forest or crop? Or do you just swap forest and crop in AFF and DEF?
lines 80-84:
Doesn’t this require additional simulations to the ones listed in the previous paragraph? These additional simulations are different than the “full expansion,”, correct?
You may want to reiterate here that the local biogeophysical responses are largely independent of the global land cover change scenario that drives the global biogeochemical response.
It appears that you don’t do this local vs non-local decomposition.
Maybe combine sections 2 and 3 into a Methods section? And relabel the included sections accordingly?
Add another section to the methods describing the creation and use (i.e. experimental design) of the tree change ssp scenarios. For example, are the changes used the changes over time from the start to the end of the scenario? Are these changes estimated from a particular model, as the scenario prescriptions do not include forest cover change?
2. Training datasets
3. Statistical emulation of temperature response
Explain why you selected January and July for your analysis.
Modeling surface temperature
lines 124-128:
Does this mean that a separate model is built for each orographical feature? This stratification needs to be clarified. How are the orographical features defined? Show some example features on a map? If these are separate models, then this should be clear in calibration, evaluation, and figure 1 also.
lines 159-162:
This isn’t clear. I assume you take the final model and run it within each block, but I don’t understand how you assign the treefrac changes and what the binning is for. Do you replace all grid cell changes within a particular bin with a specific number (as long as this number isn’t the same as the actual grid cell fraction?). Then what do you calculate the RMSE against if there is no analogue? And are you just comparing the binned RMSE with the original RMSE - what does this tell you if you don’t actually have a corresponding temperature change for the binned inputs? Do you compare these values for each bin separately? How does orographic stratification play into this?
Diagnosing 2m surface temperature
line 166:
What do you mean by observational surface temperature? What is being measured?
I assume that the observational 2m temperature is literally 2m off the ground, regardless of canopy type and height, correct?
lines 189-209:
This is a clever way to test the validity of using Tmin/max. Do you use the same gaussian kernel for both data sets? I am unfamiliar with this technique and the kernal selection seems a bit arbitrary (although you probably want it to look like a fit to each respective dataset, which indicates that the kernels would be different). In any case, you assume that each bivariate dataset for T is normal in order to use this technique. Is this the case? Did you do a regression for each dataset to check for normality and determine the appropriate kernels? Presumably you then use the each data set to train the respective covariance matrices with the respective training data error.
Emulating 2m air temperature under different land cover change scenarios
This section is misnamed, and it seems you intended to include the description of the scenario analysis.
This section needs to be moved into section 3.2 and is about uncertainty in the GAMs model.
lines 228-231:
Is this done globally for each relevant pixel? Or do you just take 200 samples total, from wherever (which doesn’t make sense if you average them)?
4. Results
Blocked cross validation results
line 248:
Please explain this month-specific approach in the methods section.
lines 261-263:
Do you mean less than 2.5? Figure 3 shows a lot of area with RMSE well over 0.25 K.
You may want to quantify how much and which areas have RMSE < 0.25 K.
An RMSE over 0.5 could actually mean that this method is not very good at representing temperature changes due to tree cover change, as such changes are often estimated on the order of 1 degree or less.
Comparing with appendix C doesn’t back up the statement about high RMSEs corresponding to extreme deforestation.
Also, based on the Obs in appendix C, extreme deforestation is realistic.
lines 265-278:
Figure 4 is just a statistical representation of figure 3, correct? Does figure 4 show the median and interquartile range? Figure 4 does show that the numbers are generally lower than they look in figure 3, in most cases. Maybe a different scale in figure 3 would be helpful.
lines 280-294:
It is unclear how you did this.
Illustration of TSmean outputs
lines 295-317:
Are these just the final models? Is the 95% interval just the distribution of the grid points within the latitudinal bands?
What about the parametric uncertainty in the model?
Surface 2m air temperature diagnosis
line 326 and figure 7:
95% is not an interquartile range
lines 327-333:
Are these results similar for the other ESMs?
Also, is this really good enough? ESMs can output more frequent temperature values, and so model values can be obtained closer to the measurement times. Maybe this improves the relationship? What are alternatives? Diurnal temperature modeling outside of the ESM? Is it possible to create a more appropriate basis using different data and Hooker’s method? I see that later you suggest model-specific fitting.
lines 336-339:
This is an overly optimistic assessment. Clearly Tmax has higher biases, evidenced by double the plot scale.
lines 340-348:
Are these results similar for the other ESMs?
5. Exploration of tree cover scenarios
These are also results and should be in the Results section.
Why did you select CESM2 for this and the other examples?
Can you compare these results to other estimates? There may be some confounding factors but there are studies estimating changes in surface temperature due to land cover change. While it may not be useful as evaluation, it certainly would be interesting to see if there are corresponding patterns.
lines 355-360:
This should be in methods.
Also, figure 9 states 2015 tree frac change. What is this actual change? Is it the change over the century?
lines 366-369:
It would be interesting to see scatter plots of signal to noise ratio vs change in tree frac. This may provide an estimate of a cover change threshold(s) that may result in significant temperature changes
6. Conclusion
lines 417-424:
This is an important point. This method does provide an opportunity to look at model-specific physical responses as it isolates the tree cover change, which is often different across models. There are still challenges related to initial conditions and model-specific climates.
lines 425-430:
isn’t there a winckler paper showing non-linearity in temperature response to change with changing initial forest fractions?
Winckler, J., C. H. Reick, and
J. Pongratz (2017), Why does the
locally induced temperature response
to land cover change differ across
scenarios?, Geophys. Res. Lett., 44,
doi:10.1002/2017GL072519.
Citation: https://doi.org/10.5194/egusphere-2022-1024-RC2 -
AC1: 'Comment on egusphere-2022-1024', Shruti Nath, 25 Feb 2023
Dear Dr. Müller, Dear Reviewers,
We would like to thank the reviewers for their time and effort in reading and reviewing our paper. We have read and considered the reviews and proposed some changes that could be made within the paper. The most important changes are:
1. We have proposed adding more details on the ESM experiments used and tree cover change ssp scenarios used2. We have included possibly adjustments to the methods to improve the chronology and understandability
3. We are more explicit in the final discussions about the outputs of the TIMBER:
a) That improvements in the 2m diagnosis are still required
b) That PFTs are not considered
c) That albedo and latent-heat fluxes would additionally have to be considered when looking at biophyiscal impacts
We are confident that the proposed responses to the reviews will strengthen the paper as well as address the concerns of the reviewers. The detailed responses are attached. We once again would like to thank the reviewers for their help, and hope that they find our responses useful.
Warm regards,
Shruti Nath
(On behalf of all co-authors)
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2022-1024', Anonymous Referee #1, 05 Dec 2022
General comments:
This manuscript describes an emulation of earth system model output, specific to the biogeophysical responses to forest cover change. Afforestation, reforestation, and reference scenarios are used to assess the model's representation of change in air temperature at and 2 metres above the land surface. The emulator appears to perform well against MPI-ESM, but substantial errors accumulate for extreme afforestation and deforestation scenarios in the other models and against observations.
The research topic is an important and relevant one, as the computational complexity of ESMs increases, as does the importance of assessing and planning land-based climate mitigation responses. The study is well-designed and clearly argued and the manuscript is well-structured and thorough. Overall I would recommend its publication with minor revisions.
Specific comments:
What about anything other than surface and 2m air temperature? At least acknowledge its limitations as a metric for all "biophysical" climate impacts (e.g. that it conflates albedo- and latent heat-related surface temperature changes despite their quite different effects on the atmospheric energy balance).
There is no mention of tree species/PFT and how the variation in tree types around the world influences biophysical properties in different biomes. Since tree PFT likely accounts for a major part of the variance of biophysical properties in the ESMs, I would have liked to know how this was translated to the emulator.
The figure captions should fully define all acronyms and variable names. It should not be necessary to read the text in detail to gain a basic understanding of what the figures show.
Technical corrections:
L5: usage: suggest "accounting for them" to replace "their accountance"
L9 and L413: usage: suggest removing "agilely", which does not feel like a natural word, and its meaning is communicated later in the sentence already (with "by means of a computationally inexpensive manner"
L57: usage: suggest "hereafter" to replace "hereon"
L59: typo: "biobphysical"Citation: https://doi.org/10.5194/egusphere-2022-1024-RC1 -
RC2: 'Comment on egusphere-2022-1024', Anonymous Referee #2, 22 Dec 2022
Review of egusphere-2022-1024:
TIMBER v0.1: a conceptual framework for emulating temperature responses to tree cover change
Summary
The authors present a new approach for estimating the local surface temperature responses to changes in tree cover. They first fit and validate multiple GAMs to model outputs and observations (separately). They also generate additional GAMs to represent parametric uncertainty in these models. They asses the validity of using an existing observational relationship to determine 2-m surface temperature from ESM estimated min and max surface temperatures. They show reliable GAM results and state that the existing relationship is good enough for this study. These tools are then applied to two SSP scenarios of tree cover change to demonstrate the utility of this approach for estimating temperature change due to change in tree cover.
Overall review
I appreciate this effort and see this as a good contribution toward understanding the potential effects of land cover change on climate. The authors employ a clever mix of statistical techniques in a thorough framework that estimates surface temperature change due to tree cover change. My main concerns are with the paper organization and incomplete description of methods. Overall the methods are sound (as far as I can tell), although one aspect in particular could be improved, and comparison with an alternative method is warranted. I recommend addressing the following three main issues. See below for detailed comments.
1) The methods are incomplete. Some methods are out of place while others are missing or unclear.
2) I am not convinced that the Hooker relationship as used is adequate for the final estimation of 2m T. For the time being, it shows the potential of this overall approach, but it seems to have a rather large spread of error for operational use. The authors should consider taking up their suggestion and doing a model-specific fit of Tmin/max to 2m T, and comparing this to the current results.
3) The scenario results should be compared with other estimates of temperature response to land cover change. There are comparable examples of this using these three ESMs. While such examples use a simulation differencing method that encapsulates much more than the method in this study, One would expect some similarities given the dominant role of tree cover change in surface temperature change. Maybe the comparison is between the TSmean change estimated by the GAMs for each scenario, rather than the 2m T, and the mean surface temperature change estimated by model differencing.
Specific suggestions and comments
Abstract
line 1:
“Society is set to experience…” is a bit overstated here. Evidence shows that the world is not even close to meeting the Paris Agreement. I suggest revising to something like “Land cover change has been proposed to play a significant role in achieving the temperature goals…”
line 6:
expand ESM
1. Introduction
lines 32-33:
you probably want to include the boysen 2020 paper in this citation also.
2. Training datasets
line 75-77:
Is there just one REF scenario? Then “The REF scenario spans 150 years…”
What do you mean by “full expansion?” Is 100% of the surface covered by forest or crop? Or do you just swap forest and crop in AFF and DEF?
lines 80-84:
Doesn’t this require additional simulations to the ones listed in the previous paragraph? These additional simulations are different than the “full expansion,”, correct?
You may want to reiterate here that the local biogeophysical responses are largely independent of the global land cover change scenario that drives the global biogeochemical response.
It appears that you don’t do this local vs non-local decomposition.
Maybe combine sections 2 and 3 into a Methods section? And relabel the included sections accordingly?
Add another section to the methods describing the creation and use (i.e. experimental design) of the tree change ssp scenarios. For example, are the changes used the changes over time from the start to the end of the scenario? Are these changes estimated from a particular model, as the scenario prescriptions do not include forest cover change?
2. Training datasets
3. Statistical emulation of temperature response
Explain why you selected January and July for your analysis.
Modeling surface temperature
lines 124-128:
Does this mean that a separate model is built for each orographical feature? This stratification needs to be clarified. How are the orographical features defined? Show some example features on a map? If these are separate models, then this should be clear in calibration, evaluation, and figure 1 also.
lines 159-162:
This isn’t clear. I assume you take the final model and run it within each block, but I don’t understand how you assign the treefrac changes and what the binning is for. Do you replace all grid cell changes within a particular bin with a specific number (as long as this number isn’t the same as the actual grid cell fraction?). Then what do you calculate the RMSE against if there is no analogue? And are you just comparing the binned RMSE with the original RMSE - what does this tell you if you don’t actually have a corresponding temperature change for the binned inputs? Do you compare these values for each bin separately? How does orographic stratification play into this?
Diagnosing 2m surface temperature
line 166:
What do you mean by observational surface temperature? What is being measured?
I assume that the observational 2m temperature is literally 2m off the ground, regardless of canopy type and height, correct?
lines 189-209:
This is a clever way to test the validity of using Tmin/max. Do you use the same gaussian kernel for both data sets? I am unfamiliar with this technique and the kernal selection seems a bit arbitrary (although you probably want it to look like a fit to each respective dataset, which indicates that the kernels would be different). In any case, you assume that each bivariate dataset for T is normal in order to use this technique. Is this the case? Did you do a regression for each dataset to check for normality and determine the appropriate kernels? Presumably you then use the each data set to train the respective covariance matrices with the respective training data error.
Emulating 2m air temperature under different land cover change scenarios
This section is misnamed, and it seems you intended to include the description of the scenario analysis.
This section needs to be moved into section 3.2 and is about uncertainty in the GAMs model.
lines 228-231:
Is this done globally for each relevant pixel? Or do you just take 200 samples total, from wherever (which doesn’t make sense if you average them)?
4. Results
Blocked cross validation results
line 248:
Please explain this month-specific approach in the methods section.
lines 261-263:
Do you mean less than 2.5? Figure 3 shows a lot of area with RMSE well over 0.25 K.
You may want to quantify how much and which areas have RMSE < 0.25 K.
An RMSE over 0.5 could actually mean that this method is not very good at representing temperature changes due to tree cover change, as such changes are often estimated on the order of 1 degree or less.
Comparing with appendix C doesn’t back up the statement about high RMSEs corresponding to extreme deforestation.
Also, based on the Obs in appendix C, extreme deforestation is realistic.
lines 265-278:
Figure 4 is just a statistical representation of figure 3, correct? Does figure 4 show the median and interquartile range? Figure 4 does show that the numbers are generally lower than they look in figure 3, in most cases. Maybe a different scale in figure 3 would be helpful.
lines 280-294:
It is unclear how you did this.
Illustration of TSmean outputs
lines 295-317:
Are these just the final models? Is the 95% interval just the distribution of the grid points within the latitudinal bands?
What about the parametric uncertainty in the model?
Surface 2m air temperature diagnosis
line 326 and figure 7:
95% is not an interquartile range
lines 327-333:
Are these results similar for the other ESMs?
Also, is this really good enough? ESMs can output more frequent temperature values, and so model values can be obtained closer to the measurement times. Maybe this improves the relationship? What are alternatives? Diurnal temperature modeling outside of the ESM? Is it possible to create a more appropriate basis using different data and Hooker’s method? I see that later you suggest model-specific fitting.
lines 336-339:
This is an overly optimistic assessment. Clearly Tmax has higher biases, evidenced by double the plot scale.
lines 340-348:
Are these results similar for the other ESMs?
5. Exploration of tree cover scenarios
These are also results and should be in the Results section.
Why did you select CESM2 for this and the other examples?
Can you compare these results to other estimates? There may be some confounding factors but there are studies estimating changes in surface temperature due to land cover change. While it may not be useful as evaluation, it certainly would be interesting to see if there are corresponding patterns.
lines 355-360:
This should be in methods.
Also, figure 9 states 2015 tree frac change. What is this actual change? Is it the change over the century?
lines 366-369:
It would be interesting to see scatter plots of signal to noise ratio vs change in tree frac. This may provide an estimate of a cover change threshold(s) that may result in significant temperature changes
6. Conclusion
lines 417-424:
This is an important point. This method does provide an opportunity to look at model-specific physical responses as it isolates the tree cover change, which is often different across models. There are still challenges related to initial conditions and model-specific climates.
lines 425-430:
isn’t there a winckler paper showing non-linearity in temperature response to change with changing initial forest fractions?
Winckler, J., C. H. Reick, and
J. Pongratz (2017), Why does the
locally induced temperature response
to land cover change differ across
scenarios?, Geophys. Res. Lett., 44,
doi:10.1002/2017GL072519.
Citation: https://doi.org/10.5194/egusphere-2022-1024-RC2 -
AC1: 'Comment on egusphere-2022-1024', Shruti Nath, 25 Feb 2023
Dear Dr. Müller, Dear Reviewers,
We would like to thank the reviewers for their time and effort in reading and reviewing our paper. We have read and considered the reviews and proposed some changes that could be made within the paper. The most important changes are:
1. We have proposed adding more details on the ESM experiments used and tree cover change ssp scenarios used2. We have included possibly adjustments to the methods to improve the chronology and understandability
3. We are more explicit in the final discussions about the outputs of the TIMBER:
a) That improvements in the 2m diagnosis are still required
b) That PFTs are not considered
c) That albedo and latent-heat fluxes would additionally have to be considered when looking at biophyiscal impacts
We are confident that the proposed responses to the reviews will strengthen the paper as well as address the concerns of the reviewers. The detailed responses are attached. We once again would like to thank the reviewers for their help, and hope that they find our responses useful.
Warm regards,
Shruti Nath
(On behalf of all co-authors)
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 581 | 173 | 15 | 769 | 9 | 12 |
- HTML: 581
- PDF: 173
- XML: 15
- Total: 769
- BibTeX: 9
- EndNote: 12
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Shruti Nath
Lukas Gudmundsson
Jonas Schwaab
Gregory Duveiller
Steven Johan De Hertog
Felix Havermann
Iris Manola
Julia Pongratz
Sonia Isabelle Seneviratne
Carl Friedrich Schleussner
Wim Thiery
Quentin Lejeune
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(6727 KB) - Metadata XML