Glacier algae accelerate melt rates on the south-western Greenland Ice Sheet

Cook, Joseph M.; Tedstone, Andrew J.; Williamson, Christopher; McCutcheon, Jenine; Hodson, Andrew J.; Dayal, Archana; Skiles, McKenzie; Hofer, Stefan; Bryant, Robert; McAree, Owen; McGonigle, Andrew; Ryan, Jonathan; Anesio, Alexandre M.; Irvine-Fynn, Tristram D. L.; Hubbard, Alun; Hanna, Edward; Flanner, Mark; Mayanna, Sathish; Benning, Liane G.; van As, Dirk; Yallop, Marian; McQuaid, James B.; Gribbin, Thomas; Tranter, Martyn

doi:https://doi.org/10.5194/tc-14-309-2020

Articles | Volume 14, issue 1

https://doi.org/10.5194/tc-14-309-2020

© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/tc-14-309-2020

© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 14, issue 1

Research article

|

29 Jan 2020

Research article |

| 29 Jan 2020

Glacier algae accelerate melt rates on the south-western Greenland Ice Sheet

Joseph M. Cook, Andrew J. Tedstone, Christopher Williamson, Jenine McCutcheon, Andrew J. Hodson, Archana Dayal, McKenzie Skiles, Stefan Hofer, Robert Bryant, Owen McAree, Andrew McGonigle, Jonathan Ryan, Alexandre M. Anesio, Tristram D. L. Irvine-Fynn, Alun Hubbard, Edward Hanna, Mark Flanner, Sathish Mayanna, Liane G. Benning, Dirk van As, Marian Yallop, James B. McQuaid, Thomas Gribbin, and Martyn Tranter

Download

Final revised paper (published on 29 Jan 2020)
Supplement to the final revised paper
Preprint (discussion started on 03 Apr 2019)
Supplement to the preprint

Interactive discussion

Status: closed

AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment

- Printer-friendly version

- Supplement

RC1: 'Comments', Anonymous Referee #1, 05 May 2019
- AC2: 'Response to Reviewer 1', Joseph Cook, 04 Jul 2019
RC2: 'Referee's report', Stephen Warren, 06 Jun 2019
- AC1: 'Response to Reviewer 2', Joseph Cook, 04 Jul 2019
- AC3: 'Revised m/s', Joseph Cook, 04 Jul 2019
  - AC4: 'Revised Supplementary Info', Joseph Cook, 06 Jul 2019

Peer-review completion

AR: Author's response | RR: Referee report | ED: Editor decision

ED: Reconsider after major revisions (further review by editor and referees) (08 Jul 2019) by Marco Tedesco

ED: Publish subject to revisions (further review by editor and referees) (07 Aug 2019) by Marco Tedesco

AR by Joseph Cook on behalf of the Authors (07 Aug 2019) Author's response Manuscript

ED: Referee Nomination & Report Request started (08 Aug 2019) by Marco Tedesco

RR by Stephen Warren (21 Aug 2019)

Suggestions for revision or reasons for rejection

This paper is still not ready for publication. Concerning the melting ice surface of the West-Greenland ablation zone in summer, the authors argue that addition of dust causes the ice albedo to increase, so that any reduction of ice albedo by light-absorbing impurities would be due to algae. This denial of a role for dust in reducing albedo is based on several questionable arguments, which I will point out in this review. Most of these questionable arguments can be classified into one of three classes of disconnects: (a) The authors’ response to my review, saying how the paper was changed, does not correspond with what the revised paper actually shows. (b) The way figures are described in the text does not correspond to what the figures actually show. (c) References cited in support of a claim do not actually provide the claimed evidence.

I do not doubt that algae absorb sunlight, but I do doubt the quantitative attribution of albedo change on Greenland to algal abundance. A lot of work went into this project, and I would like to see it come to more robust conclusions. It has the potential of becoming an important paper.

Major comments

(1) The major comment of my first review was to point out that the imaginary part of the dust refractive index should decrease across the visible spectrum from 300 to 700 nm, not increase as shown in the first submission. In their Response to Reviewers, the authors now thank me for “pointing out the unrealistic refractive index”, but then in their revised manuscript their dust imaginary index still increases across the visible (by a factor of 2.7 from 300 to 700 nm), as shown in the new Figure 3C. [An example of Disconnect Type “a”.]

(2) I don’t want this extraordinary claim of blue dust to enter the literature on the composition of Greenland ice without further evidence. What mineral composition gives it the blue color indicated by Figure 3C? In their response to my review, the authors indicate that the mineralogy of the dust is consistent with Figure 3C, and that a paper on this topic is soon to be submitted by one of the authors (McCutcheon). A brief summary of the mineralogy in that forthcoming paper should be included in this paper; it could be cited as “unpublished data” or “manuscript in preparation”.

(3) The authors have ignored the request in my Major Comment #1, in which I asked the authors to show the computed albedo effect of the measured dust concentration; they continue to show just the computed albedo effect of arbitrary amounts of dust (100, 300, 500 ppm), and similarly for arbitrary concentrations of algae. At least a table is needed, giving measured dust and algal concentrations (ppm by mass). The numbers of cells are shown in Figure 2C, but these need to be combined with algal-cell size distributions to get the mass.

(4) Lines 272-273. Tedesco et al. (2013) is cited as indicating “a lack of red mineral phases”. In fact, Tedesco’s Figure 6a shows that both dust and algae are “red”, and that dust is redder than algae; Tedesco speculated that the goethite they found in their samples had dehydrated to hematite in the drying and heating process. Goethite is the hydrated form of hematite; it is not as absorptive as hematite but its absorption coefficient likewise decreases across the visible. It is true that goethite has a yellow appearance rather than red, but it is misleading to cite Tedesco as finding “a lack of red mineral phases”, since the present authors are using the adjective “red” here to characterize the spectral slope of reflectance, which increases toward the red for both goethite and hematite. [Disconnect type “c”]

(5) Lines 518-521. “The imaginary refractive index of the mineral dust sample (Fig 3C) . . . indicating . . . scarcity of red minerals in the bare ice.” This comment is not forthright; there is no mention of the factor-of-2.7 increase of imaginary index from 300 to 700 nm, which indicates not merely a scarcity of red minerals but actually a dominance of blue minerals. Don’t be so timid! You must highlight this strange imaginary index, and point out how it contradicts the behavior reported by Tedesco et al. 2013. [Disconnect Type “b”]

(6) Lines 539-541, discussing Figure 3B. Albedo spectra for algae “downsloping with increasing wavelength between 0.35 and 0.45 microns . . . and a gentle increase to 0.70 microns. These spectral features are consistent with our field spectra for algal ice” This is not true. The field spectrum for algal ice (Figure 2B) starts at 0.40 not 0.35, and shows albedo increasing, not downsloping, from 0.40 to 0.45 microns. And Figure 2B (field) shows a steep increase from 0.6 to 0.7 microns, whereas the dashed line in Figure 3B (model) is flat from 0.6 to 0.7. [Disconnect Type “b”]

(7) Lines 574-575. “. . . algal cells had a greater albedo-reducing effect than mineral dusts in north-west Greenland (Aoki et al. 2013).” This summary of the Aoki paper is misleading. Aoki et al. did conclude that the imaginary index of algae was larger than that of mineral dust, but did not conclude that most of the albedo reduction was due to algae. Their total impurity mass in the ice was 1127 ppm, of which 29 ppm (2.6%) was organic carbon (algae), so ~1100 ppm was dust. Their Figure 4b shows that they could explain most of the albedo reduction by 1000 ppm dust; the remainder (which looks like about 5% to me) is then attributed to algae. [Disconnect Type “c”]

(8) Lines 544-546, and Figure 2C. The authors point out that mineral dust particles can “act as substrates for the formation of low-albedo microbial-mineral aggregates”. This suggests an alternative explanation of the correlation shown in Figure 2C: The algae may be concentrated in patches of ice that have high mineral content, so the cell count then would be correlated with the albedo reduction caused by dust.

(9) Section 3.3, lines 577-610. This section analyzes the albedo trough centered at 1.02 microns, following Nolin and Dozier (2000). Nolin and Dozier found the 1.02-micron trough to be deeper for lower albedo (coarse-grained snow), whereas Supp Info 5B here shows the opposite, namely deeper trough for higher albedo. The 1.02-micron feature is therefore not useful for discussing “indirect effects of algae”. The entire section 3.3 should therefore be shortened to just the last four lines 607-610, making a single sentence starting “Algal growth is stimulated . . . “ [Disconnect Type “c”]

(10). Figure 4ABC. Half of the solar energy is at wavelengths <0.7 microns, and 80-90% (depending on cloud thickness) is at wavelengths <1.0 microns. In these figures, the peculiar wiggles in the visible region, the most energetically important part of the spectrum, are squeezed into a tiny region on the far left of the figures. These wiggles need to be discussed and explained in the text, and the figure should be redrawn, for a domain 0.3-1.5 microns instead of 0-5 microns.

Minor comments.
line 206. “. . . they are large, far outside the domain of Mie scattering”. Mie theory is not restricted to small size-parameters. Admittedly Mie calculations do become expensive for large size-parameters.

line 247. “real refractive index”. I think you mean imaginary refractive index, or complex refractive index.

line 262. “four global average dusts from Flanner et al. (2007)”. The Flanner 2007 paper concerns black carbon; it contains no information about dust. Give the correct reference. [Disconnect Type “c”]

line 320. “CI” has not yet been defined.
line 478. Change 2B to 2D.
line 484. Change 2C to 2B.
line 488. Change 2C to 2D.

Figure 2B. The labels Hbio, Lbio, CI, SN should be defined in the figure caption.

Figure 6. This is a nice figure showing the contrast between 2016 and 2017.

Table 2 should be rearranged. Seeing the numbers from different methods side-by-side would help the reader compare them. For example, the first line (water albedo) would read “0.31, 0.08, 0.08”.

Spelling and punctuation.
line 181. Change gluteraldehyde to glutaraldehyde.
line 290. “the the”
line 983. distributed
line 1240-1245 (Table 1). “ug” is not the correct symbol for micrograms. Use the Greek lower-case mu.

Hide

RR by Anonymous Referee #1 (25 Aug 2019)

Suggestions for revision or reasons for rejection

In this revision, Cook et al. provided more details on methods as suggested by reviewers. The fieldwork and radiative transfer modeling components are very important and valuable for the cryosphere community to understand better the impurity properties. However, some key/core information/figures are still lacking for the section using UAV and Sentinel-2 remote sensing data assisted by ASD field spectra to conduct image classification. In addition, there are some contradictory and inconsistent arguments throughout the text. These problems need to be resolved:
1) I asked the authors to provide the ASD spectra (resampled to UAV bands and Sentinel-2 bands) compared against the real UAV and Sentinel-2 spectra, but I didn’t see them in this revision. I think the authors should at least provide a figure showing how consistent the ASD resampled spectra are with the UAV and Sentinel-2 spectra. This is very important for the image classification, because UAV spectra and Sentinel-2 spectra have uncertainties of atmospheric correction and other radiation correction procedures. If they don’t match very well, using ASD-resampled spectra to train the RF or other classifiers is not convincing. Although the authors claim that the RF method is advanced and novel for studying the cryosphere, it is essentially an image classification problem and subject to basic image classification rules. The authors classified surface types into snow, clean ice, cryoconite, water, low biomass algae, and high biomass algae. Please provide a figure showing the comparison between the ASD reduced spectra and the UAV and Sentinel-2 spectra for each class, and label the bandwidth for each UAV and Sentinel-2 band. If the spectral curves of these classes are highly correlated, the whole classification problem will become a thresholding problem, and the derived ‘high biomass algae’ will be equivalent to ‘dark ice’. Without providing this information (i.e. actual plots), the classification method is not solid, and the classification results are not convincing.
2) The arguments regarding glacier algae and red mineral dust are contradictory. In the introduction section, when the authors mentioned the recent study by Wang et al. (2018) who proposed to use the reflectance ratio between 709 nm and 673 nm to map ice algae in southwest Greenland, it is claimed that ‘the red-edge is potentially vulnerable to false positives due to red mineral dusts (Seager et al. 2005; Cook et al. 2017b)’. On page 8 line 260-265, the authors mentioned again that red dust has ‘similarly shaped spectral albedo to glacier algae’. However, throughout the paper and the two referenced papers, I didn’t find any convincing evidence (i.e. data) showing that the red dust can indeed mask out the chlorophyll signature at red-NIR spectrum. If the red dust is indeed very similar to glacier algae in spectral properties, then no practical methods would exist to separate them regardless of the spatial resolution of the remote sensing data, which means the RF method cannot work as well. The authors claim later in the manuscript (on page 14 line 500-510) that the chlorophyll-a signal between 670 and 710 nm (red edge) is uniquely biological and use this red-edge (Painter et al., 2001 and Wang et al., 2018) to support their results. On page 16 line 550-555, the authors again argue that red minerals occur only in very low concentrations and would have a negligible effect on the ice optics. These arguments are contradictory, with the later arguments supporting the red-edge method used for mapping algae in southwest Greenland. I suggest the authors should properly acknowledge previous studies.
3) In the response, the authors said ‘We attempted to use the vegetation red-edge on our spectra and achieved only ~80% accuracy for identifying algal ice compared to our random forest classifier that achieves >95% accuracy.’ How many samples were used to calculate the accuracy? Were the training samples also used to obtain the 95% accuracy? If they were, it is not a fair comparison. Actually, for remote sensing classification problems, 80% accuracy is quite high. Why not use this information to cross-validate your method? It seems to me that the red-edge method actually supports your classification results.
4) Page 7 line 227-228, please show simulation results in supplementary information.
5) Page 7 line 376, does this mean that there are 231*0.7=162 training samples and 231*0.3=69 test samples? Were all accuracy tests based on 69 samples? In this case, how many samples for each class were available to test the classifiers?
6) Page 11 line 387-390, I don’t quite understand the logic here, are the authors trying to say that they want to build an endmember library using ASD spectra? How can it be certain that using ‘pure’ spectra can get better accuracy for classifying images (with spatial heterogeneity within each pixel) rather than directly labeling the images (to obtain training sets)?
7) Supp info 6: this figure is not easy to read, please directly give the correlation numbers in a table format, along with the number of test samples.
8) I’m confused by your study area. On page 12 line 420, “The area of interest was the “common area” defined by Tedstone et al. (2017) bounded within the latitudinal range 65 – 70 N, and is equal to that used by Wang et al. (2018)”. Figure 1 and Figure 5 show an area between 67.615N and 67.599N, which is much smaller than 65 – 70 N. Which one is the study area? It seems that the UAV and Sentinel-2 only cover a very small portion.
9) Section 2.10 and 3.7. Thanks to the authors for providing more details here. However, this part is still not very clear to me. When the authors generalized the runoff estimate (simulated based on three weather stations) over the entire region (65 – 70 N?), are the area percentages of low biomass algae and high biomass algae based on classification results over the 200*200 m UAV region and the 67.615-67.599N Sentinel-2 region to generalize to the entire region based on the bare ice area and dark ice area derived from MODIS data (500 m resolution)? If so, did the authors consider issues associated with different spatial scales? The authors pointed out that ‘The higher detection limit for algae with decreasing ground resolution makes our estimate of spatial coverage from Sentinel-2 conservative. We highlight that this will have a much larger effect on studies aiming to quantify cell abundance using Sentinel-3 where the ground resolution is 300 m.’ (page 19 line 669). Table 3 shows the area percentage of different classes, total algae is 78.5% on UAV, 57.99% on Sentinel-2 (2016) and 58.87% on Sentinel-2 (2017). Is this the evidence that supports the argument that Sentinel-2 estimate is conservative? If so, this doesn’t make sense since totally different areas are being compared (200*200m vs 10000*10000m). Please clarify. Given the spatial scale issues, it is incorrect to directly extrapolate the area percentage from centimeter resolution and 20m resolution to MODIS 500 m resolution. Please justify the methods.

Hide

ED: Reconsider after major revisions (further review by editor and referees) (26 Aug 2019) by Marco Tedesco

AR by Joseph Cook on behalf of the Authors (26 Sep 2019) Author's response Manuscript

ED: Referee Nomination & Report Request started (18 Oct 2019) by Marco Tedesco

RR by Anonymous Referee #2 (06 Nov 2019)

RR by Anonymous Referee #1 (12 Nov 2019)

Suggestions for revision or reasons for rejection

This manuscript is not suitable for publication on The Cryosphere. The authors barely incorporated my comments into their manuscript. Given the coarse spectral resolution of the UAV data they are using, I am not convinced that the glacier algae can be separated from other dark materials using Random Forest method, since they can all suppress the visible spectrum, making the whole classification downgrade as a thresholding problem (although your random forest may learn a multi-threshold to do the classification). This is shown by figure 3 and figure 4 in the response report (the algae-covered area is just darker). These two figures are very difficult to read without exact wavelengths marked on the x-axis. To me, the ASD, UAV and Sentinel-2 spectra in these two figures do not match in terms of shape. Essentially if the remote sensing data do not contain algae-unique information, no matter what kind of machine learning or artificial intelligence algorithm you are using, it would not work by using insufficient datasets, since algorithms themselves cannot generate new information. This is also the reason why glacier algae could not be mapped previously. In your last round revision, you used the band ratio method by Wang et at. (2018) and compared it with your classification results. I remember you can get pretty good cross-validation results, and I suggested to use that method (already published) to support your method. Apparently, the authors are very confident in their classification method by using only 45 test points. The so-called ‘red-edge’ method (by the authors) is a very sophisticated model used by ocean color studies for many years to retrieve chlorophyll-a content in aquatic systems. It is very misleading that the authors keep saying it is ‘simply a presence/absence detector’.

Hide

ED: Publish subject to minor revisions (review by editor) (05 Dec 2019) by Marco Tedesco

AR by Joseph Cook on behalf of the Authors (05 Dec 2019) Author's response Manuscript

ED: Publish subject to minor revisions (review by editor) (15 Dec 2019) by Marco Tedesco

AR by Joseph Cook on behalf of the Authors (16 Dec 2019) Author's response Manuscript

ED: Publish as is (18 Dec 2019) by Marco Tedesco

AR by Joseph Cook on behalf of the Authors (18 Dec 2019) Manuscript

Short summary

Melting of the Greenland Ice Sheet (GrIS) is a major source of uncertainty for sea level rise projections. Ice-darkening due to the growth of algae has been recognized as a potential accelerator of melting. This paper measures and models the algae-driven ice melting and maps the algae over the ice sheet for the first time. We estimate that as much as 13 % total runoff from the south-western GrIS can be attributed to these algae, showing that they must be included in future mass balance models.