Landsat-Based Estimation of the Glacier Surface Temperature of Hailuogou Glacier, Southeastern Tibetan Plateau, Between 1990 and 2018

Round 1

Reviewer 1 Report

The study by Liao et al presents a time-series of land-surface temperatures for Hailuogou Glacier derived from Landsat observations using a mono-window algorithm for 1990-2018, and examines the multidecadal trend in surface temperatures.  This is an interesting analysis and the first study I am aware of that has examined long-term glacier surface temperature change, and has the potential to be a nice contribution.  At present, however, the study seems to be a first-level presentation of results, and needs considerable revision and additional analysis before being suitable for publication.  The introduction and results could be more clearly framed, and the validation effort is poorly justified considering the multi-sensor, multidecadal, multi-seasonal scope of the study. Very little analysis has been done to correct for the aliasing of their necessarily nonuniform satellite observations to likely seasonal patterns, although the station data shows considerable promise for this purpose. As a result, the presentation of the principal results (the trends) are very messy at present, and do not even provide goodness-of-fit metrics, seriously undermining the strength of the results. More careful consideration of the uncertainty of the method is needed, as well as the strength of the trends in glacier surface temperature, and the manuscript would benefit from additional analyses of surface temperature changes in the debris-covered areas and the accumulation areas, for example. In addition, the discussion of implications of the results is essentially lacking. In short, the manuscript needs additional work, but has the potential to be a very nice contribution to Remote Sensing.

Main points:

Considerable English-language editing is needed by a specialist.  The quality of the written English gets in the way of readability, especially in the introduction.

Presentation of methods could be clearer in places. How are the suitable scenes chosen? How are clouds controlled for? How will the scenes be compared to one another (what metrics)? What does the glacier mean surface temperature indicate?

The validation effort is not very thorough, and not justified well.  The authors compare the results of their method to two other methods for 1 scene from 1 season (using L8).  First, it would be best to validate with independent data. This may not be possible, but that needs to be made clear. Then, these alternative methods (which also disagree) need to be demonstrated to be sufficiently accurate that they can be used for validation. Are they all consistent terms of the improvement relative to simply using the brightness temperature? I am a bit doubtful! Then, the mono-window method needs to be shown to work consistently for all seasons and for both sensors analysed.  This is a considerable effort, but is important to justify the determination of long-term changes in glacier surface temperature without on-ground validation. How confident are the authors that there is no bias between OLI and TM sensors, for example? What about the possibility of seasonal biases for the method?

The presentation of results is not very tidy or convincing at present – very simple line plots of quite noisy data with nonuniform sampling.  Some additional postprocessing could really improve the presentation of the results and the strength of the conclusions. How strong are the trends? What does the seasonal signal of surface temperature look like (not its trend but the seasonal pattern)?

The attribution of causality for the observed changes in glacier surface temperature is also weak, because little analysis has gone into debris, topography, and climate as drivers. For the debris cover, it would be worthwhile to actually derive multitemporal debris coverage maps and related indices, as well as the mean surface temperature within the debris-covered area. For topography, one could model topographic shading using any of many available models, then correlate this with the anomaly to your elevation-gradient fit (equivalent to the red line in Figure 5, but for all scenes). For air temperature, this appears to be a clear proxy to the mean glacier surface temperature, and additional analysis is needed to demonstrate and test this relationship.

Key comments.

L24. Please include a rate here – how quickly is this changing?

L31. ‘of the earth land’ is an odd phrase. Do you mean of soil/vegetation as opposed to water, or…?

L32-33. Should be ‘at regional and global scales’

L43. Would be nice to reference some examples here

L51-52. It is not clear here what improvements were made by [14] to the model of [10].

L56. ‘Unfortunately,’ seems to missing an entire sentence here?

L67. inadequate in what way?

L73. How does the comparison with other algorithms constitute a validation? It needs to be made clear why this demonstrated independent validation, rather that multiple algorithms produce the same results when supplied with the same data…

L83. ‘entire glacier’, not ‘entail glacier’

L88. 2 km in 200 years is not very remarkable as a retreat rate, except that most debris covered glaciers do not retreat much. This should be made clearer.

L90. These flow velocities are indeed remarkable.  At what position is the 41m/a (how far from the terminus?)

L100. None of these data were acquired at 30m resolution, although this is the resolution that USGS/NASA produce them at. This needs to be made clear. Also, USGS/NASA need an acknowledgement.

L102. ‘A total of 99 images…’

L104. What do you mean that ‘all images were selected seasonally’? Just that they were classified by season? Or selected to represent all seasons? Or…? Some more details are needed regarding how images were selected – based on cloud cover? Temporal coverage? Some other quality metric? How did you choose 1 best scene for each season, for example?

Table 1 could go into an appendix or supplementary information

L115. Here it seems important to give the reader some sense of structure. Maybe add a line stating that you will now explain how you derive all the inputs needed for Equation 1.

L152-165. So if I understand it, you use the emissivity of ice for your entire domain? In this case there is no need for Table 5.  However, as you note that there is debris cover, you should consider if your results would change much for that domain using a distinct emissivity (for example).

L172. Why ‘intend to’? Did you or not?

L189. Can you combine Tables 7 and 8? Also, there is a typo in the Table 8 caption.

L197. I suppose that H is the elevation (in m? km?)? These equations are all from Yang et al [41] or a different source?

L205. I do not quite understand how this is meant to be a validation – these methods also have their own uncertainties and other problems. Furthermore, you validate the results with 1 scene, representing a single sensor and season. If this comparison is meant to be sufficient to convey confidence in the results, it needs to be extended seasonally and to the L45 sensor. Are the biases constant across different inputs and seasons? This is vital to ensure that the trend is meaningful. At present, I am not convinced. I know that validating LST estimates is a huge challenge, but this is a very minimal effort, especially for a multidecadal effort.

Figure 2. Which of the three algorithms do you trust the most, and why? You need to be able to come up with an overall uncertainty, or for each scene, or even for each pixel.

L223. This should be ‘RTE’

L224. Before results, it would be worthwhile to have a very short section describing how you will analyse your results! I.e. derivation of the glacier mean temperature, and the 0C isotherm elevation. Other useful metrics might be the extent of the glacier below 0C or below -1C (to reflect the area that is at or above the melting point), and the mean temperature in the debris-covered area, etc.

L235. The determination of this line may be somewhat sensitive to the choice of emissivity. In particular, the accumulation area and ablation area will each have multiple emissivities (snow, ‘ice’, debris) varying in spatial coverage. Some assessment of uncertainty of the 0C isotherm elevation is worth including (if you have an algorithm uncertainty of 0.5C, for example, how uncertain is the isotherm?

Figure 3. This is a nice depiction of the results for one scene! It would be great to include a similar depiction for different seasons and for the earlier Landsat data, which will help to convince me that the algorithm is reliable for this whole period. Also, a nice way to depict the spatial heterogeneity would be to show the anomaly of surface temperature from the altitudinal fit (ie show in space the difference between the black dots and the red line).  This would highlight the magnitude of variations.

Figure 4. It would be nice to show all the results for each season in the background as grey lines (not indicating the year, just a spaghetti plot to show that all the profiles follow this general pattern), then to include these 4 examples for each season coloured as you have them.

L249. The decreasing trend in summer is not apparent in Figure 4b. It looks like the 2017 summer surface temperatures are the warmest yet.

L253. The increasing trend of the 0C isotherm doesn’t reflect the ablation area extent, but it reflects the debris-covered area, since theoretically only the debris-covered area should go above the melting point!

Figure 5. The linear fits here do not appear to explain much of the variation in LST. It would be very good to put an R2 and p-value on each subplot to highlight the confidence in a trend.  I think part of the difficulty here is that 1) trends could be different in ablation and accumulation areas, 2) glacier mean surface temperature doesn’t have a very strong physical meaning, and should probably be separated into debris-covered and debris-free areas, 3) there is a big risk of aliasing because only 1 scene per season is used for each year.  Perhaps it would be useful to plot glacier mean surface temperature with respect to month, to see if the seasonal pattern is resolved more clearly? Other metrics would be very interesting to highlight, such as the portion of the glacier that is at/near the melting point?

Figure 6. This is quite noisy, because it includes the seasonal variations, but due to data availability the seasons are not sampled uniformly (some years have <4 seasons, but also the date changes year to year for any season). It would be much more useful to 1) show the seasonal GST evolution for all years (x axis is day of year or month, etc), 2) derive a fit to this to estimate the mean annual GST  and apply it over all the years – this would give a much tidier comparison between years. You could do the same with the 0C isotherm elevation.  Both panels also need R2 and p-values, as well as the trend equation.

L266. I’m quite surprised to see cloud cover arising in the discussion, but not before. Did you select scenes based on cloud cover at all, or all all scenes cloud-free over the glacier itself? In any case, the authors are absolutely correct that clouds in the scene (over the glacier or not) will indicate a higher likelihood of atmospheric water affecting the rest of the scene.  In addition, cloud-shadows can lead to erroneous LST retrievals and reduce comparability between scenes. This absolutely needs to be filtered out in scene selection.

L274. What do the authors mean by ‘glacier surface dark’? This section is very brief, and it would be very good if the authors put in the effort to show the debris-covered area expansion based on their selected scenes. Perhaps a more important question is whether the glacier surface itself is warming, or if the glacier area is warming on average simply because the debris-covered area is expanding.  Both could be occurring, and the debris-covered area could also be warming (further biasing the signal from what the rest of the glacier is experiencing). For this reason, I would really focus on distinct domains for this postprocessing.

L276. Malting->melting

Figure 7. Why is a Sentinel-2 scene used here, rather than the L8 from Figure 8? No Sentinel-2 data have been mentioned before this point. Also, this sort of comparison is difficult because the scene dates do not correspond to one another – the 1990 image is much earlier in the ablation season, so it is not surprising that it shows more snowcover! In addition, the terminus retreat is not evident; this could be helped by indicating the 1990 terminus position in subsequent figures for comparison.

Figure 8.  Can this be combined with Figure 7?

L295. I’m not sure what you mean by ‘redistributing the radiation’. The following two points make sense (restricting illumination, emission of longwave radiation), but I don’t think this really qualifies as a redistribution of radiative energy…

L301. Thus ‘exhibits’ lower surface temperature in most seasons.

Figure 9.  This is quite a convincing plot at first glance, but how it was derived is not entirely clear.  For the air temperature, were these derived based on precisely the scene data and time, or are they mean daily temperatures for the day of the satellite observation, or what? Perhaps more importantly, this relationship looks strong enough that the Mt Gongga Station observed air temperature could be used as a proxy for mean glacier surface temperature (this should be checked with regression). If this is the case, I would prefer to see the seasonal and annual temperature analyses performed on this dataset, since it is more robust as a long-term, uniformly-spaced set of observations. The discussion of different patterns and trends across the glacier of course makes sense with the spatial data.

L319. What are the implications of this study? This should be discussed! If these trends continue, how will the annual 0C isotherm elevation look in 50 years? How much of the glacier will be experiencing melt year-round? What does this mean?

L330. The gradient is per km, correct? Please indicate units.

L331. I’m not convinced about debris cover in the accumulation area, this looks like exposed bedrock within the glacier outline.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

English needs some smoothing in places. The manuscript would benefit from English editing and should be checked for typos and placement of commas in particular.

The `Study area` is briefly presented and I think more details regarding the description of the geology, relief, climatic characteristics (temperature and precipitation distribution, maybe a graph), ELA should be added here.

The figures need to be improved (see specific comments in the attached material).

In the abstract and conclusions, you mentioned that you compared the results of the mono-window algorithm with other algorithms, but within the Results/Discussions, there are no comparisons offered. In my opinion, you should compare your model with other existing to show if this new technique brings any improvement. This comparison is highly needed for this study to meet the standards for this journal. You need to elaborate more discussions on why this model is a better option than previous ones.  

There are no validation data for your model and, thus it isn`t easy to assess the quality of your model. Is there any GST data at the meteorological station nearby?

The clouds can induce some uncertainties in your results. Can you somehow present the mean annual GST together with clouds coverage within the contour of the investigated glacier? and to elaborate a discussion on this problem?

It seems that the GST curve follows pretty nice the air temperature variations. However, it is not clear to me if in Figure 9 you represented mean annual temperatures, seasonal temperatures or means monthly temperatures?

Specific comments are attached!

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

First, I hope that the authors and their families are safe and healthy right now.

Very interesting study looking at glacial temperatures. Overall fairly well done, but extensive work needs to be corrected on the English and Grammar. See comments ordered by Line Number.

Was there overlap between 5 and 8 data to compare temps from same date.location? Any difference from Sensors?

If emissivity is altered slightly from 'bare ice' to a value representing it is debris covered, how does this alter the results? The ice is not really 'perfectly' bare ice - it is dirty and altered.

I did not check the accuracy of the intext citations and the Cited List. Please make sure that those are accurate.

Some headings used all capitalisation, some did not. Check for consistency.

Line# Comments:

11 - 'influenced'

31 - 'Earth's'?

33 - 'scales'

35 - 'research'

35 - 'retrieval'

36 - 'source'

37 - 'research'

38 - 'recognized as an indicator'

39 - 'occurring'

41 - Remove 'Due to .. surface,'

41 - 'influenced'

54 - Remove 'There being', start at 'A variety ..'

56 - Unfortunately hanging word?

57 - 'has limited investigation'

64 - Remove 'Besides,'

68 - 'concentrated'

69 - 'amount of images'

73 - Remove 'retrieving'

74 - Replace 'reasonability' with 'accuracy'?

80 - Replace 'altitude' with 'elevation'

81 - Replace 'zoon' with 'zone'

83 - 'entail'? 'entire'? Not sure here.

84 - 'mantle' wrong word here? 

89 - 91 - Flow velocity is increases further from tongue? Is that correct? Or a mis-word.

92 - FIG1 needs lat / long for geolocation.

102 - Remove 'Totally' replace with 'In total'

109 - TABLE1 - Remove the underscore '_' from the Table Labels to allow them to be shown multi-line and not broken. See Date Aquired.

128 - 'retrieval'

155 - 'simulate'

158 - 'derive'

159 - 'calculate'

197 - 199 - the font for equations 9 to 11 have been altered, making them appeared stretched.

237 - is this not the ELA line? Used in other Glacial Studies.

238 FIG 3 - the numbers and values shown in (b) are very hard to read.

242 - FIG 4 - this figure would have a better impact if the y axis were the same - makes it easier to compare and the lines would match up better.

248 - 'temperatures on Hailuogou...'

249 - '.. while a slight decrease occurred ..'

254 - please make all axis equal for comparison. It looks like 'Date' label has been cut off on first two. Might want to add letter labels a), b), c), d), to match and be consistent with other Figures in paper.

263 - 265 - Reoword this sentence, it is a run-on and should probably be broken into 2.

265 - Replace 'essential' with 'important'

266 - 'as follows: (1) the .. study; (2) it is .. field; and (3) natural surfaces ..'

283 - what is the rate of upward migration of the debris on the glacial surface?

287 - 288 - watch extra spaces around reference and periods.

295 - Remove 'on'

296 - 298 - Reoword this sentence, it is a run-on and should probably be broken into 2.

313-314 - 'and an increasing trend'

317-318 Caption colour definitions do not match actual figure data. They appear to be reveresed?

324 - Remove 'which makes the method applicable easily'

327 - Why not compare model temps to the ground station near by? apply modelled lapse rate for temp change and see if they are close?

330-331 - negative 6.12degC per what distance? a gradient is a rate, not a single value.

333 - Replace 'the application' with 'this study'

334 - 'the ablation area is increasing since...'

335 - 'temperatures'

 

Author Response

Please see the attachment.

Author Response File: Author Response.doc

Round 2

Reviewer 1 Report

Please see the attachment

Comments for author File: Comments.pdf

Author Response

Major comments:

1. I understand that the validation is quite difficult, and appreciate that you have extended the comparison to multiple seasons and all sensors used. This is a big improvement for assessing the confidence of the results, even without in-situ data. A few aspects need to be stated more clearly in the text. For example: which of the three methods do you consider to be‘best’ in the absence of ground-truth data? You seem to disregard the RTE estimate as less accurate. Do you think the disagreement between methods sufficiently represents the uncertainty of the LST retrieval, or is it possible that all three methods are biased for this high-mountain environment? This deserves a bit of discussion in the main text. You clearly show a considerable change in LST in all seasons exceeding this uncertainty, which certainly strengthens your results (the uncertainty is probably smaller than the change). However, I would still suggest to frame this comparison a validation, but an uncertainty assessment of your results. In this respect, the outcome is clearly that different methods would product roughly the same result (an increase in temperatures). 

Yes, for the satellite derived land surface temperature, their validation with ground true data is almost impossible, because of the exceptionally large difference between point measurements and gridded pixel-based (usually with tens or hundreds of meters spatial resolutions) estimations. This study mainly aims to show how seasonal and interannually dynamics of glacier surface temperature of a partly debris-covered glacier, by using a selected land surface estimation from long term satellite observations. Since the uncertainty is probably smaller than the change, we still could draw some conclusions about how the changing climatic and glacier surface geomorphologic processes, etc., have impacted on the surface thermal conditions of a debris-covered glacier. A possible solution will be employing thermal camera sensor, which could be mounted on some UAVs, to mapping surface temperature of the glacier over a enough larger area. That kind of distributed measurement would be possible used as a better validation for the satellite observations.

 

2. The linear trend analyses (Figs 7-9) are still not convincing. These trends should be computed accounting for seasonality and the nonuniform sampling in your satellite data. This is a long-recognized problem in hydrology (e.g. Hirsch et al, 1982; Shao and Li, 2011; Anghileri et al, 2014) that has seen attention in remote sensing as well (e.g. Eastman et al, 2009; Verbesselt et al, 2010). A simple trend of all the data is really not appropriate or acceptable for this analysis. Using the appropriate regression in this respect will provide much more convincing, interesting and useful results, and a much more-frequently cited output. As this is the first effort to implement this analysis for a mountain glacier (note that Eastman et al 2009 shows LST trends for the entirety of the Greenland Ice Sheet), it seems vital to assess this trend in a robust and convincing manner.

Yes, we realized that the simple line trend analyses are still not convincing. In our revised manuscript, we replaced these simple trend line analysis with a more statistically presentation. In revised Fig.7, we now just present the temporal variation of the station record air temperature and satellite derived glacier surface temperature. In revised Fig. 8, we at first compared the values between station record air temperature and satellite derived glacier surface temperature, then we use box plots of each 5 year statistics for the station air temperature, the glacier wide temperature and the debris-covered area surface temperature. In this way, we could show that surface temperature of the glacier wide mean and on the debris-covered area have experienced warming. The warming trend on the debris-covered area is much more obvious than the glacier wide mean (so also than the debris-free parts). We also did the same analysis for the 0℃ isothermal line elevation and extent of the glacier surface above 0℃ in the revised Fig. 9.

 

Minor comments:

Thank you for clarifying methodological details of scene selection and emissivity, and for isolating the debris-covered area changes, etc. Also, the seasonal altitudinal profiles are a very nice addition to the manuscript. 

L24. Debris thickening has not been demonstrated directly in the manuscript, only debris cover expansion.

We replaced the ‘thickening’ with ‘expanding’.

L56 ‘has seen limited investiagtion’

Revised.

L68 ‘researches’ should be ‘research’ or ‘investigations’

Revised as ‘investigations’.

L73. Simplify ‘and its application to’ to ‘for’

Revised as your suggestion.

L84. ‘Rock debris covers XX…’

Revised as ‘Superficial debris covers …’. Because debris also include some fine materials in addition to rock.

L105. ‘provided by the USGS at 30 m…’

Revised.

L179. ‘We used a band ratio approach to…’

Revised as ‘We used a band ratio approach to eliminate the impact of topography and enhance the discrimination between different snow grains.’

L226-227. ‘yielded at a bias’ sounds strange. Better to say simply ‘exhibited’ or ‘have’

Revised.

L250. It looks like the text has accidentally been formatted as a heading

Thanks, here we revised the wrong text format.

L263. This comma should be a semicolon

Revised.

Figure 5. Please clarify in the caption that these are the mean or median for each elevation

The ‘mean’ is added in the caption.

L317. You mean that it would be good to validate the surface emissivities you have estimated, right? This could be stated more clearly. How large would the effect of an incorrect emissivity be for your LST estimates?

Yes, any ground in situ point measurements cannot offer validation for the gridded pixel’s merged value. But in our revised manuscript, we have employed K-Mean unsupervised classification-based ice/snow type for different emissivities, which as far as the best choice for ice/snow surface determination.

L319. ‘Glacier surface dark’ is really not a widely used term. I think it would be better to specifically identify black carbon and/or cryoconite (if relevant), as these terms are generally recognized

Here we revised as: ‘Impact of Glacier Debris Cover expanding and Ice Surface Darkening on GSTs.’

L327. This could be written more clearly. BC and dust increasingly on the glacier surface reduce glacier albedo, increasing net shortwave energy and resulting in higher surface temperatures.

We revised the text with your suggestion.

Figure 11 does not provide much additional information and should be integrated with Figure 10.

As suggested by the 2^nd reviewer, now we added contours inside the glacier and also included darkening areas for each period.

 

Hirsch, R. M., Slack, J. R., and Smith, R. A. ( 1982), Techniques of trend analysis for monthly water quality data, Water Resour. Res., 18( 1), 107– 121, doi:10.1029/WR018i001p00107.

Shao, Q., & Li, M. (2011). A new trend analysis for seasonal time series with consideration of data dependence. Journal of hydrology, 396(1-2), 104-112.

Ronald Eastman, Florencia Sangermano, Bardan Ghimire, Honglei Zhu, Hao Chen, Neeti Neeti, Yongming Cai, Elia A. Machado & Stefano C. Crema (2009) Seasonal trend analysis of image time series, International Journal of Remote Sensing, 30:10, 2721-2726

Verbesselt, J., Hyndman, R., Newnham, G., & Culvenor, D. (2010). Detecting trend and seasonal changes in satellite image time series. Remote sensing of Environment, 114(1), 106-115.

Anghileri, D., Pianosi, F., & Soncini-Sessa, R. (2014). Trend detection in seasonal data: from hydrology to water resources. Journal of Hydrology, 511, 171-179.

Reviewer 2 Report

I am happy with the revised version of the manuscript and I recommend the publication of this paper in Remote Sensing.

However, there are some minor issues that need to be addressed before publication:

Line 254: the title of the subchapter should be placed on the next line;

Lines 258 and 259: You provide values for temperature gradients, but I think these values correspond to 100 m, not to 1000 m.   Air temperature decreases generally with 0,6 degrees / 100 m and 6 degrees / 1000 m.

Figure 10 looks very nice, but the scale bar is missing.

Figure 11. Please, also insert scale bar. It would be useful to insert main topographical contours within the polygons representing the glacier extent. 

Author Response

I am happy with the revised version of the manuscript and I recommend the publication of this paper in Remote Sensing.

Thank you agian for offering us further suggestions.

However, there are some minor issues that need to be addressed before publication:

Line 254: the title of the subchapter should be placed on the next line;

Revised.

Lines 258 and 259: You provide values for temperature gradients, but I think these values correspond to 100 m, not to 1000 m.   Air temperature decreases generally with 0,6 degrees / 100 m and 6 degrees / 1000 m.

Yes, this should be our mistakes. Revised.

Figure 10 looks very nice, but the scale bar is missing.

Now the scale bars are added.

Figure 11. Please, also insert scale bar. It would be useful to insert main topographical contours within the polygons representing the glacier extent.

We added contours inside the glacier and also included darkening areas for each period.