the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Improving modelled albedo over the Greenland ice sheet through parameter optimisation and MODIS retrievals
Abstract. Greenland ice sheet mass loss continues to accelerate as global temperatures increase. The surface albedo of the ice sheet determines the amount of absorbed solar energy, which is a key factor in driving surface snow and ice melting. Satellite retrieved albedo allows us to compare and optimise modelled albedo over the entirety of the ice sheet. We optimise the parameters of the albedo scheme in the ORCHIDEE land surface model for three random years taken over the 2000–2017 period and validate over the remaining years. In particular, we want to improve the albedo at the edges of the ice sheet since they correspond to ablation areas and show the greatest variations in runoff and surface mass balance. By giving a larger weight to points at the ice sheet's edge, we improve the model-data fit by reducing the RMSD by over 25 % for the whole ice sheet for the summer months. This improvement is consistent for all years, even those not used in the calibration step. We conclude by showing which additional model outputs are impacted by changes to the albedo parameters encouraging future work using multiple data streams for optimisation.
-
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
(1087 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(1087 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2022-745', Anonymous Referee #1, 17 Oct 2022
- AC1: 'Reply on RC1', Nina Raoult, 02 Dec 2022
-
RC2: 'Comment on egusphere-2022-745', Anonymous Referee #2, 22 Oct 2022
This article presents the calibration of the ORCHIDEE model against the MODIS derived snow albedo dataset. While the overall objective of improving albedo is very relevant, this particular study, in my opinion, is very limited. My specific concerns are outlined below.
The article reads like a description of the research in the way it was conducted. The authors describe all the methodologies the authors tried, which are sometimes distracting from the main objective of the paper. For example, Section 3.2. describes the results with two different optimization algorithms. As shown here (and as well known), gradient search methods have limitations in exploring complex decision spaces within an optimization context. The results presented here are not adding anything new to the key focus of this paper, and it is distracting. In Section 2.4.2 – It is not clear (at this point) in the manuscript what is meant by ‘performing a sensitivity analysis of the model’. Typically, this is done ahead of the calibration step to reduce the number of parameters being optimized (as the authors acknowledge in Section 3.4.3). If that’s the same context, it’ll be good to describe that and present this section before 2.4.1. Similarly, Section 3.4.3 should be presented earlier (even though that’s not how this work evolved). I appreciate the value of explaining all the steps, but there are lots of ‘preliminary’ setups (section 3.2, line 207) in this paper. A major recommendation is to restructure the paper so that it focuses on the finalized results, while presenting the intermediate results and steps only to support the main findings.
As the authors note in the summary, calibration has its problems in that adjusting certain model parameters may improve some parts of the model, while degrading others. The main objective of improving albedo is to improve the changes in the snow pack over GrIS, as noted in the intro. The paper needs to describe what the impact of the improved snow albedo formulation is on the snow simulations (and other model states). Does the improved albedo lead to better snow states?
The modeling setups use forcing data from MAR, which is a modeled estimate, presumably with its own associated biases and errors. In a calibration setup, the tuned parameters are then used as an error sink to ‘hide’ these boundary condition errors. This needs to be discussed in the article. Is there an evaluation of MAR data over GrIS? Are other ‘observational’ datasets available?
Since ORCHIDEE is used in global setups, how are the results over this domain applicable in a general sense? Are these calibrated parameters limited to GrIS?
Minor comments:
Line 51. Need brackets around Krinner et al. (2005).
Line 67: Change to ‘the’ instead of ‘our’?
Section 2.3 – It is important to clarify (here, early on in the paper, abstract, and title) that the snow albedo is being calibrated instead of the total albedo. MODIS has several different albedo products (blue-sky, black-sky etc.) Please clarify.
Line 146: change to ‘output’ instead of ‘writing’
Line 151: remove ‘However’
How do the calibrated values influence the peak winter month simulations?
Figure 1 – this is the snow covered albedo? Is this average computed by excluding Nov-Feb?
Section 3.1 – This is a very hand-wavy section. The authors need to spell out exactly what was changed in this manual calibration procedure. What parameters/physics were changed?
Section 3.2: How many iterations of GA were used here? Are these the results from the ‘Both’ approach (results in Figure 2)?
Table 2: How are the albedo evaluated for ‘All months’? If you don’t trust the MODIS albedo during the winter months, how do you justify comparing back to them?
Line 220: Why were these three years chosen? How do you do these calibrations (separately for each year and somehow harmonize the calibrated parameters? Or are they calibrated from a single run, but the calibration data is withheld during all years except 2000, 2010, and 2012)?
Line 223: Add a comma after ‘Indeed’.
Line 232: Why is it that ‘We would not expect to lower the RMSD of the edges any further’?
Citation: https://doi.org/10.5194/egusphere-2022-745-RC2 - AC2: 'Reply on RC2', Nina Raoult, 02 Dec 2022
-
EC1: 'Comment on egusphere-2022-745', Patricia de Rosnay, 04 Nov 2022
Dear authors,
Thank you for your submission to The Cryosphere. While both reviewers recognise the merit of your paper major revisions are required before it can be published to The Cryosphere. This includes improved clarity in the description of the MODIS product used, as pointed out by Reviewer 2. Also Section 3 requires to be slightly re-organised to avoid giving intermediate steps that were conducted to explore different minimisation approaches. Focusing to the actual approach will clarify the main messages of the paper.
Best wishes,
Patricia de Rosnay
Citation: https://doi.org/10.5194/egusphere-2022-745-EC1
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2022-745', Anonymous Referee #1, 17 Oct 2022
- AC1: 'Reply on RC1', Nina Raoult, 02 Dec 2022
-
RC2: 'Comment on egusphere-2022-745', Anonymous Referee #2, 22 Oct 2022
This article presents the calibration of the ORCHIDEE model against the MODIS derived snow albedo dataset. While the overall objective of improving albedo is very relevant, this particular study, in my opinion, is very limited. My specific concerns are outlined below.
The article reads like a description of the research in the way it was conducted. The authors describe all the methodologies the authors tried, which are sometimes distracting from the main objective of the paper. For example, Section 3.2. describes the results with two different optimization algorithms. As shown here (and as well known), gradient search methods have limitations in exploring complex decision spaces within an optimization context. The results presented here are not adding anything new to the key focus of this paper, and it is distracting. In Section 2.4.2 – It is not clear (at this point) in the manuscript what is meant by ‘performing a sensitivity analysis of the model’. Typically, this is done ahead of the calibration step to reduce the number of parameters being optimized (as the authors acknowledge in Section 3.4.3). If that’s the same context, it’ll be good to describe that and present this section before 2.4.1. Similarly, Section 3.4.3 should be presented earlier (even though that’s not how this work evolved). I appreciate the value of explaining all the steps, but there are lots of ‘preliminary’ setups (section 3.2, line 207) in this paper. A major recommendation is to restructure the paper so that it focuses on the finalized results, while presenting the intermediate results and steps only to support the main findings.
As the authors note in the summary, calibration has its problems in that adjusting certain model parameters may improve some parts of the model, while degrading others. The main objective of improving albedo is to improve the changes in the snow pack over GrIS, as noted in the intro. The paper needs to describe what the impact of the improved snow albedo formulation is on the snow simulations (and other model states). Does the improved albedo lead to better snow states?
The modeling setups use forcing data from MAR, which is a modeled estimate, presumably with its own associated biases and errors. In a calibration setup, the tuned parameters are then used as an error sink to ‘hide’ these boundary condition errors. This needs to be discussed in the article. Is there an evaluation of MAR data over GrIS? Are other ‘observational’ datasets available?
Since ORCHIDEE is used in global setups, how are the results over this domain applicable in a general sense? Are these calibrated parameters limited to GrIS?
Minor comments:
Line 51. Need brackets around Krinner et al. (2005).
Line 67: Change to ‘the’ instead of ‘our’?
Section 2.3 – It is important to clarify (here, early on in the paper, abstract, and title) that the snow albedo is being calibrated instead of the total albedo. MODIS has several different albedo products (blue-sky, black-sky etc.) Please clarify.
Line 146: change to ‘output’ instead of ‘writing’
Line 151: remove ‘However’
How do the calibrated values influence the peak winter month simulations?
Figure 1 – this is the snow covered albedo? Is this average computed by excluding Nov-Feb?
Section 3.1 – This is a very hand-wavy section. The authors need to spell out exactly what was changed in this manual calibration procedure. What parameters/physics were changed?
Section 3.2: How many iterations of GA were used here? Are these the results from the ‘Both’ approach (results in Figure 2)?
Table 2: How are the albedo evaluated for ‘All months’? If you don’t trust the MODIS albedo during the winter months, how do you justify comparing back to them?
Line 220: Why were these three years chosen? How do you do these calibrations (separately for each year and somehow harmonize the calibrated parameters? Or are they calibrated from a single run, but the calibration data is withheld during all years except 2000, 2010, and 2012)?
Line 223: Add a comma after ‘Indeed’.
Line 232: Why is it that ‘We would not expect to lower the RMSD of the edges any further’?
Citation: https://doi.org/10.5194/egusphere-2022-745-RC2 - AC2: 'Reply on RC2', Nina Raoult, 02 Dec 2022
-
EC1: 'Comment on egusphere-2022-745', Patricia de Rosnay, 04 Nov 2022
Dear authors,
Thank you for your submission to The Cryosphere. While both reviewers recognise the merit of your paper major revisions are required before it can be published to The Cryosphere. This includes improved clarity in the description of the MODIS product used, as pointed out by Reviewer 2. Also Section 3 requires to be slightly re-organised to avoid giving intermediate steps that were conducted to explore different minimisation approaches. Focusing to the actual approach will clarify the main messages of the paper.
Best wishes,
Patricia de Rosnay
Citation: https://doi.org/10.5194/egusphere-2022-745-EC1
Peer review completion
Journal article(s) based on this preprint
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
412 | 188 | 18 | 618 | 1 | 4 |
- HTML: 412
- PDF: 188
- XML: 18
- Total: 618
- BibTeX: 1
- EndNote: 4
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1
Sylvie Charbit
Christophe Dumas
Fabienne Maignan
Catherine Ottlé
Vladislav Bastrikov
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(1087 KB) - Metadata XML