Publisher’s note: this comment is a copy of AC5 and its content was therefore removed.

Publisher’s note: this comment is a copy of AC5 and its content was therefore removed.

Publisher’s note: this comment is a copy of AC5 and its content was therefore removed.

Publisher’s note: this comment is a copy of AC5 and its content was therefore removed.

Many thanks for this very thorough and constructive review. Please find our responses below. We have numbered the items for referencing and are limiting the response to items highlighting potential issues and stating specific questions. We provide additional details in several cases including the grouping of the CAVM (item 9). This includes a clarification of the group names. Figures  are provided in a supplement (pdf).

1)

The main problem with this study, is that it is difficult to understand the units in the map and how they would best be used in future analyses. Because the unit characteristics are summaries of ground data points, there is a lot of overlap in moisture, organic soil depth, and vegetation cover characteristics. The problem with this mapping approach is seen when one considers the difficulty of carrying out an accuracy assessment. It may not be possible to determine if a ground point is correctly mapped when distinctions (cut points) between the units are not well quantified or described.

• Reply: Note that the derived statistics represent to some extent geolocation uncertainties (in situ and satellite) and are partially limited in informational value due to the low number of samples. The soil and vegetation statistics are presented separately but should be interpreted together (as was done for the naming of the units). For example, volumetric water content statistics appear similar for many units, but their vegetation patterns differ.  
• The overlap also exemplifies the heterogenous Arctic landscapes. It documents how far we can get with a 10 m nominal resolution. The 10 m cannot fully resolve landscape patterns, but provides a step forward. The statistics itself (standard deviation, box plots) could be used as an indicator for heterogeneity within the units.

2)

Grouping the units into broader, more distinct categories can help, but eliminates much of the benefit of the fine spatial resolution. The title and abstract suggest that this mapping might be most useful for mapping moisture gradients (4 units: aquatic, wet, moist and dry). However, the legend categories are not clearly differentiated by soil moisture characteristics as seen in Figure 4, and one particularly problematic unit (#8) is characterized in Table 2 as aquatic to dry (but < 1% of map). Figure 6 shows that for the Western Siberia area, all map units had all four levels of moisture, though some were dominated by one or two moisture levels. Unit #6, which is listed in Table 2 as dry is shown is Figure 6 as having mostly moist ground data.

• Reply: Note that the target moisture units are 'wet', 'medium' and 'dry' (see grouping B in Table 2). The mentioned 4 units are used in the vegetation in situ dataset (describing relevés, Figure 6). Figure 4 shows data from a soil dataset (point data). Both are considered for the descriptions. The reference to figure 6 is actually missing in line 288, and the text not directly referring to it what probably led partially to the confusion. In addition to correcting this we propose to switch the presentation of the in situ statistics (first releves and than soil data).
• We agree that the word ‘represented’ in the abstract, line 5, is misleading. The intention was to give a technical detail, the sampling of 10 m, and not the ability to capture the gradients fully at 10 m. Indeed, the spatial resolution of 10 m cannot fully resolve moisture gradients. Unit #8 represents landscapes with very high heterogeneity regarding wetness. However, as pointed out, this unit occurs only over 1% of the analyzed domain. A further challenge is the assignment of in situ points to certain pixels/units. Uncertainties exist for the determination of GPS coordinates in the field and also regarding geocoding of the satellite data since acquisitions with different viewing geometries (optical and SAR) are combined. The assignment of the four wetness units for the releves in the field is also partially subjective. These issues together result in assignments of ‘wet’ in situ points to dry units such as #6 and #7. But the proportion is comparably low. The sum of ‘aquatic’ and ‘wet’ in situ points is in the order of 20%. In case of the listing as wet in Table 2 it is >50%. In case of #6, in situ soil moisture data include also comparably dry sites (Figure 4, <20% volumetric content, comparably high number of samples). This led to the decision for assignment to group ‘dry’, what should have been explained in more detail. The comparison to the Alaska landcover map (Figure 3 below, item 8) shows that #6 primarily occurs in upland and alpine landscapes. Also the comparison with CAWASAR (Figure 1 below, item 3) confirms #6 as dry. However, we propose to discuss the impact of the decision to label #6 as dry by adding a second wetness heterogeneity calculation (with #6 as moist).
• We further propose to rephrase the abstract:
• Old: ‘Patterns of lakes, wetlands, general soil moisture conditions and vegetation physiognomy are represented at 10 m nominal resolution.’
• New: ‘Patterns of lakes, wetlands, general soil moisture conditions and vegetation physiognomy are interpreted at 10 m nominal resolution.’
• Old: ‘The result goes beyond the capability of existing landcover maps which have deficiencies in spatial resolution, thematic content and accuracy.’
• New: ‘The result goes beyond the capability of existing landcover maps which have deficiencies in spatial resolution, thematic content and accuracy although landscape heterogeneity related to moisture gradients cannot be fully resolved at 10m.’

3)

If the main purpose of the map is for wetlands/soil moisture, I would have expected it to be compared to the CAWASAR (Circum Arctic Wetlands based on Advanced Aperture Radar) map (Obu et al., 2019). Instead, the map is compared to three land cover maps (CCI, CALC-2020, and Raster CAVM), using 6 grouped categories. One problem with this approach is that the other map legends have to be cross-walked to the 6 groups, and in the case of the Raster CAVM, that was not done correctly (see more detailed comments below).

• Reply: Please find our comments on the CAVM grouping in the detailed responses below (#9). We also provide a comparison to CAWASAR (originally published as Widhalm et al. 2015) in Figure 1 below. There is specifically agreement regarding the CAWASAR class 'dry' with the CALU group 'dry' (classes #6 and #21). The distinction between wet and moist is less reflected (scale issue?).

Supplement Figure 1: Comparison of CALU (10m, for legend of unit IDs see Bartsch et al. 2023) with CAWASAR wetness levels (500m, Widhalm et al. 2015) over the extent of the Alaskan North Slope as defined in Muller et al. (2018) (see also Figure 3).

4)

It could be that the meaning of the legend units would best be defined on a local or regional scale, rather than the circumpolar scale. The mapping seems to work quite well in the training and adjacent validation and example areas. However, the paper does not present maps of widely different parts of the Arctic, except in a small circumpolar map in Figure 11A. The extrapolation of the classification from the training area in Western Siberia to the circumpolar scale is one of the most questionable aspects of the method, so it needs to be shown to be effective, perhaps by comparing this CALU mapping with detailed local maps, where they are available for specific areas – perhaps from Long-Term Ecological Research (LTER) areas in the Arctic.

• Reply: We have now prepared comparisons with regional maps from Alaska and the Lena Delta in the supplement to this response. However, the issue with such maps is that they are usually based on 30 m Landsat data. So most pixels are expected to be composed of different units except for comparably homogeneous regions such as the high Arctic with barren grounds and permanent snow/ice. We had therefore not included such comparisons. But we agree that it would be helpful for assessment of the method transfer from Western Siberia to the entire Arctic. Note, that the used in situ soil probes represent sites across the Arctic, including Alaska, Canada and Greenland.

5)



Lines 121-123 – It is important to document the location of the in-situ (ground) data, so users of the map can evaluate the strengths/weaknesses of the map in their particular area. A map of ground data locations should be included in the Appendix.

• Reply: We agree.

6)

I suggest eliminating the word “transect” in favor of “region”, as the training region and validation region are not linear transects (Figure 1 caption and map, and text). I got confused with these different terms. Line 221, 250 – change “transect” to “training area”

• Reply: We agree.

7)

Lines 221 -228  - Please describe the extrapolation of the prototype map to the entire circumpolar area in more detail. You say the “units remained largely the same”. I would like to see this documented in some way in the Appendix. The use of this method, based on ground plots in only Western Siberia, to map the entire Arctic, is the weak point of this project. It is important to show that this did not prove to be a problem to convince people that this mapping is useful on a circumpolar scale.

• Reply: We agree that the phrasing “the units remained largely the same” is too vague. The largest change was change in the order. We agree that it would be useful to add a table in the appendix to document this re-assignment.
• New (line 224): ’19 of the 21 original classes have been assigned to the new units. The remaining two classes were modified and split up into two units each. First, the original (needle leave) forest classification was split regionally (bioclimatic zone E versus A to D) to address misclassification of tundra as forest. The disturbance unit originally combined different disturbance types which lead to vegetation removal, geomorphological processes as well as fires. A k-means classification was applied targeting two classes. Results were preliminarily assessed with data from recently burned areas within the prototype extent. The originally three water units have been merged into one water unit. Further on, a snow and a shadow unit have been introduced based on training data from Svalbard resulting in eventually 23 units. The modified prototype classification was then used to train the maximum likelihood classifier.’

8)

Lines 252-255 – The validation of the map relies on comparisons with other maps. You have done this by grouping the other map legends into 6 types. Reducing the number of legend units generally increases accuracy, so provides little information about the accuracy of the 21 CALU units. 

• Reply: As suggested above, we have now made comparisons with ‘ungrouped’ regional maps (Lena Delta and Alaska). 

Supplement Figure 2: Cross-comparison of CALU (10m) with the Lena Delta Landcover classification by Schneider et al. (2009), 30m (proportion of CAL Unit within class; for legend of unit IDs see Bartsch et al. 2023).

Supplement Figure 3: Cross-comparison of CALU (10m) with the Alaska landcover/ecosystems 2010 classification (30m) published in Muller et al. (2018) (proportion of CAL Unit within class; for legend of unit IDs see Bartsch et al. 2023).

9)

There is also a problem in choosing how to group the legends of the other maps. In my opinion, as primary author of the Raster CAVM, the CAVM units are not grouped correctly in Appendix C-2. The Raster CAVM does not map most of the Arctic as shrub dominated. As a first response, I would say that types B3, B4, G1 and P1 should be in the barren group, B2a should be lichen/moss. G3 and G4 should be graminoid (but not grassland, as they are sedge-dominated). I do not know the CCI Landcover map, as it is unpublished, and cannot judge the groupings. The CALC-2020 map has few enough types that they seem to match the broad groupings well. I suggest that the authors work with the people who created the comparison maps to get the best possible crosswalk between the map units.

• Reply: We agree that the naming of the groups is confusing in the current version. The groups for comparison do not fully reflect botanical categories specifically regarding shrublands, the used ‘shrub tundra’ is misleading. We therefore propose to change the naming to ‘dwarf to low shrub tundra’ for clarification. Considering this change, regarding G3 and G4, which have dwarf shrub according to the CAVM legend, we would keep it in the ‘dwarf to low shrub tundra’ category. But we agree for B3 and B4. A similar naming issue exists for ‘barren’. Our ‘barren’ category excludes occurrence of graminoids. Thus B2a, G1 and P1, which have graminoids according to the CAVM legend would need to be kept separate. But instead of ‘grassland’ the category needs to be renamed ‘Graminoids’ (as is also suggested by you below) and the barren category needs to be clearly explained in the caption and text.

10)

Table 2 - Change grassland to graminoid (and in appendix C and related figures). Grass-dominated areas are uncommon in the Arctic; sedge-dominated areas are common. At the 10-m resolution, grasslands would occur in revegetated disturbed areas, but this does not seem to be the definition of CALU #6, and would be rare, scattered pixels – probably not enough to make a unit on the circumpolar scale.

• Reply: We agree to rename the group grassland to graminoids.

11)

The low area for snow/ice indicates that you eliminated the Greenland Ice Cap and perhaps other year-round snow/ice. This should be described in the methods.

• Reply: We agree

12)

Section 4.4 - The text says “Burned areas from events before 2014 are represented through other classes (vegetation recovery).”, however the map in Figure 10 (right) shows areas with fires in 1997 mapped as unit #17. Please clarify.

• Reply: Figure 10 shows a special case where fires occurred several times over the same area. The most recent one overlapped several old ones, but the extent was not clear with the chosen type of visualization (with year label in polygon centre). We propose to use a different line type for the most recent fire, what then clearly shows that the disturbed class occurs only within this boundary (Figure 4 below)

Supplement Figure 4: Example for class 17 ’recently burned or flooded’ (shown in red). Burned areas in southwestern Alaska (outlines and years from Alaska Fire Database, https://www.frames.gov/catalog/10465). Dashed lines – boundaries of fires before 2019, bold line - fire extent 2019. (for legend see Bartsch et al. 2023)

13)

“The disturbance unit #17 occurs within recently burned areas in approximately 72 % of all cases.” How do you know this if you only have fire data from Alaska?

• Reply: We agree that the phrasing is misleading.
• New (line 329): ‘ … of all cases documented for Alaska.’

14)

Section 4.6 – The start of this section should include a summary of what is shown in Figures 15 and 16.

• Reply: We agree.

15)

Section 4.6.1 - Please add to the text a summary of how much the overall proportions of mapping units changed, and a comparison matrix (like Table 4) in the Appendix.

• Reply: We agree that it would be useful to add the Table in the appendix and additional text.

16)

Section 4.6.3 – As described in my comments above under Methods, I do not think the comparison shown in Figure 17 and 18 are helpful due to the problem with the grouping. The maps shown in Figures 15 and 16, of areas in southern Western Siberia, are reasonable comparisons, but are all in Western Siberia.

• Reply: The choice for Western Siberia was determined by the extent of the prototype. But we agree that also other regions should be included.

17)

Section 4.6.4 – Lines 370-375 and Figure 18 – the comparison between CALC-2020 and CAVM is not very helpful here, as readers are trying to understand CALU, not these other two maps. I can understand the utility of this comparison for the authors, but I suggest moving Figure 18 to the Appendix, and focusing the text in this section on the comparison between CALC-2020 and CALU.

• Reply: We agree to move it to the appendix

18)

The Discussion Section needs work. It does not address the major issues in this study. It should discuss the effectiveness of the classification method used, in expanding from a training area to the circumpolar area. It should describe the effects, benefits, drawbacks of using a legend derived from statistically modeled groups (this study) vs. a legend of units defined by plant community composition or moisture units or soil organic content. The existing text should be rewritten with separate paragraphs summarizing the CALU’s effectiveness in mapping diversity, wetlands, and soil organic content. Comparisons to other maps should be minimal (moved to the Results 4.6)

• Reply: We agree to the change in focus and restructuring

19)

Line 135 – clarify the “analysis extent”. The training region, the evaluation region, the entire Arctic?

• Reply: We refer here to the entire Arctic,

20) 

Line 212 - What does “Up to eight granules have been considered” mean? Considered for what? Do you mean “Up to eight Sentinel-2 granules (acquisition dates) were selected for any particular location”?

• Reply: Yes, it refers to the (fixed size) regular granules (tiles) of Sentinel-2. Their boundaries are shown in Figure 2 of the manuscript.

21)

Line 249 – “Classes have been grouped for comparison due to the large differences in thematic content” Explain this sentence more clearly. Which classes were grouped? For comparison with what? Which classes had large differences in thematic content? What types of differences?

From looking at the supplemental material and Table 2, I would say “A simplified set of nine units was developed that allowed cross-comparison between the CALU and the three external maps. Table 2 shows the grouping for the CALU, and Appendix C shows the groupings for the external maps.

• Reply: Yes, your suggested phrasing clarifies it.

22)

Line 276 – explain how you identified “anomalous meteorological/hydrological conditions”

• Reply: We used re-analyses data (as used for Sentinel-1 scene selection, line 216) to assess if a year was unusually dry or wet in summer.

23)

Line 277 - explain what is “too late or too early in the season”

• Reply: It refers to days at the beginning and end of the scene selection period (mid July to mid August) and varying seasonality/phenology between years.
• New: ‘Low quality is usually caused by acquisitions close to mid-July and mid-August in years with deviations in phenology (early or late spring).’

24)



Line 298 – please describe, 48% of what? 48% overall average cover? Occurs in 48% of plots?

• Reply: average coverage

25)

Lines 304-305 – you say that CALU #2 was not represented in the AVA, yet it is shown in Table 2 as having 51 samples.

• Reply: The used AVA dataset includes a number of plots which have been taken along lake margins, also at the boundary to unit #2. Due to geolocation uncertainties in the in situ data as well as satellite retrievals, some are located in unit #2 pixels. A typical example is shown in Figure 5 below.
• We propose to remove the line from the table to avoid confusion.

Supplement Figure 5: Example for an AVA vegetation plot (Zemlianskii et al. 2023) located along a lake margin (square), falling into unit #2 (light blue, abundant macrophytes, shallow lake margin) instead of land area due to geolocation uncertainties. (for legend see Bartsch et al. 2023)

26)

Line 314-315 – tell us which CALU are you referring to as “permanent wetland class” (#3, I think) and “seasonal wetland class” (#4) and explain what measure has contrasting low and high standard deviation (it’s not water volumetric water or mineral content based on Figures 4 and 5)

• Reply: Thanks for spotting. Some references are wrong, and a table got lost when moving to the appendix. The table reference in the first line should be Table A2 instead of Table 3. The table with the standard deviation was moved to the appendix, but is actually not included in this version of the manuscript.

27)

Line 316 – you say only 6 pedon samples were available for CALU #16, but doesn’t Fig. 8 show that 21 were available?

• Reply: This refers to sites with full soil description for which only 6 are available (see figure B50). Figure 8 shows the sites with soil organic layer thickness what is available from more locations. There is actually a ‘)’ missing after ‘A2’ on line 316 what may have led to the misunderstanding.

28)

Lines 325-326 – change “The assignment to the disturbance unit #17 does occur up to four years” to “Pixels were assigned to the disturbance unit #17 for up to four years”. The text says this, however the map in Figure 10 (right) shows areas with fires in 1997 mapped as unit #17. Please clarify.

• Reply: See our clarification above (number 12). The fire 2019 overlapped several preceding fires.

29)

lichen/moss (Table 11).” Maybe change to “Figures 15 and 16), with 63% shrub tundra, about 9% grassland and 16% lichen/moss CALU grouped units.” There is no Table 11. Would you like to include more tables in the Appendix?

• Reply: Thanks for spotting. This is a remainder of an earlier version of the manuscript. It should refer to Table 4 and the shrub fraction is 60% instead of 63% (and sum 85%).

30)

Line 387-389 – This last sentence makes a separate, important point that should be expanded into a full paragraph, perhaps moved to lines 404-410.

• Reply: This is indeed an important point. We agree to expand the paragraph.

31)

Lines 392-393 – “over the top meter” – does this mean on the surface of the ground or soils to 1 m depth? Please clarify.

• Reply: It refers to soils to 1m depth

32)

Figures 4, 5, 8 captions – add description of horizontal axis, including definition of CAL Unit, the meaning of the numbers at the top of the figure, and the meaning of the boxes, lines, plus and circle symbols

• Reply: The description of the horizontal axis is included in all cases (‘CAL Unit’). For better understanding we propose to extent it to ‘CAL Unit ID’.

33)

Figure 7 – this figure is not necessary. The data would be better presented by adding as “SD” column to Table 3. Possibly the same for Figure 12.

• Reply: We propose to move to a table if rotated tables are possible (is otherwise too wide).

34)

Figure 14 – missing legend. I cannot tell what the different shades of gray mean. Once the legend is in there, the caption will need to explain it. I originally took the comma to mean a decimal-place marker. A separate legend would be clearer.

• Reply: The legend is included and explained in the caption (‘sum of groups ...’). The confusion comes probably from the comma which separates the legend item from the actual value. We propose to remove the value so that only the legend entry remains and to extent the caption to ‘sum of groups … expressed in numbers 0 to 3.’

References

Bartsch, A., Efimova, A., Widhalm, B., Muri, X., von Baeckmann, C., Bergstedt, H., Ermokhina, K., Hugelius, G., Heim, B., and Leibmann, M.: Circumarctic landcover diversity considering wetness gradients, EGUsphere [preprint], https://doi.org/10.5194/egusphere-2023-2295, 2023.

Muller, S., D.A. Walker, and M.T. Jorgenson. 2018. Land Cover and Ecosystem Map Collection for Northern Alaska. ORNL DAAC, Oak Ridge, Tennessee, USA. https://doi.org/10.3334/ORNLDAAC/1359

Schneider, J., G. Grosse, and D. Wagner. 2009. “The Lena River Delta - Land Cover Classification of Tundra Environments Based on Landsat 7 ETM+ Data and Its Application for Upscaling of Methane Emissions.” doi:10.1594/PANGAEA.759631. In Supplement to: “Land Cover Classification of Tundra Environments in the Arctic Lena Delta Based on Landsat 7 ETM+ Data and its Application for Upscaling of Methane Emissions.” Remote Sensing of Environment 113 (2): 380–391. doi:10.1016/j.rse.2008.10.013.

Widhalm, B., Bartsch, A., and Heim, B.: A novel approach for the characterization of tundra wetland regions with C-band SAR satellite data, International Journal of Remote Sensing, 36, 5537–5556, 2015.

Zemlianskii, V., Ermokhina, K., Schaepman-Strub, G., Matveyeva, N., Troeva, E., Lavrinenko, I., Telyatnikov, M., Pospelov, I., Koroleva, N., Leonova, N., Khitun, O., Walker, D., Breen, A., Kadetov, N., Lavrinenko, O., Ivleva, T., Kholod, S., Petrzhik, N., Gunin, Y., Kurysheva, M., Lapina, A., Korolev, D., Kudr, E., and Plekhanova, E.: Russian Arctic Vegetation Archive—A new database of plant community composition and environmental conditions, Global Ecology and Biogeography, https://doi.org/10.1111/geb.13724, 2023.

Many thanks for your valuable comments! Please find our response below.

Comment: … it could be still argued, is now used thematic division relevant enough, some classes does not look very sensible, and how well they in the end are mapped.

• Reply: With our study we present how far it is possible to go on Arctic scale with current remote sensing capabilities. The thematic division represents the separability of Arctic landscapes using satellite data as it is based on unsupervised clustering (k-means). Target classes have not been predefined. The division also represents the potential of 10m nominal spatial resolution which is clearly not sufficient to fully address landscape heterogeneity in many cases. A further challenge is the interpretation of the clusters/units which relies on extensive consistent in situ observations. We agree that both needs to be better explained and discussed.

Comment: Used classification algorithm is not described well at all (rows 117-9). It seems that it is described in more detail in refereed paper, but I would added here some more information. For example, what was criteria to select just those 25 classes?

• Reply: Since the approach is based on k-means, no target classes have been predefined. The resulting classes are units of similar reflectance of the shortwave incoming radiation and radar backscatter at C-VV. We agree that this needs to be better described.

Comment: Then, was there any way to handle/smooth the mixed-pixel effect, like some fuzzy method using nearest cells, etc – and if not, I would discuss would it have been sensible to try to use some?

• Reply: The prototype was developed at 20m nominal spatial resolution as also the 20m-bands of Sentinel-2 have been used to better exploit the spectral capabilities. In this study a super-resolution approach adjusted to Arctic environments was now applied to bring these bands to 10m. The approach is based on convolutional neural networks (see line 205). The documentation of the units with in situ data clearly shows the limitation of the nominal 10m. The inclusion of the 20m bands widens the spectral resolution but impacts actual spatial resolution. We agree that it should be better discussed what the implications of the super-resolution processing are. We propose to extent the discussion (line 379) with a detailed presentation which specific units have changed between the prototype and the new version (currently this is only presented for grouped classes).

Comment: Used field vegetation data coming from Russian Arctic Vegetation data Archive, covers only European (mostly) Russian and west Siberian region. I could argue that in other locations in the Arctic, there are vegetation communities not matching well communities found in this region. This data set was harmonized, and it was also from same area, where earlier version of land cover classification using similar method was done, so its use was in that sense understandable. However, when the goal is make really relevant circumarctic product, in the next phase more different data sets other part from the arctic should be also included. This issue is shortly mentioned in rows 410-11, but I would, like to see some more discussion, how this regional “bias” might have affected the product. This then also might lead to some extent different thematic content of the product.

• Reply: We agree that the discussion should be extended regarding this issue. However, note that the soil data represent sites across the entire Arctic. We propose to add a map of their distribution for clarification.

Comment: Some thematic classes looks to some extent odd, like classes 6 and 8, where both wet and dry habitats are included. This is problem, when thinking, for example, methane dynamics. The logic behind selected thematic classes could be explained and justified still better. And some adjustments would be good to do, when next version of this product (hopefully) is produced. However, I well understand huge challenges to try to formulate such general classes, that could cover all circumarctic vegetation communities and other land cover types, and could still be mapped from remote sensed data.

• Reply: As noted above, the 10m are insufficient to fully resolve  Arctic landcover heterogeneity. This is specifically the case for units #6 and #8, which are each areas spectrally in one class but represent a range of moisture conditions. What needs to be also noted in this context is the assignment of the names based on the in situ data statistics. There are uncertainties in geolocation of both the satellite product and the in situ data. The representation of the in situ description for the 10x10m is also an issue. This mismatch may also lead to the fuzzy/broad naming of the units in heterogeneous areas.

Comment: I was to some extent surprised to notice that organic layer thickness was so low in those classes that looks like paludified, like classes 3, 4, 5 (and 6?), and how it was so high in class 15 (see Figure 8)? This could be at least discussed.

• Reply: Thanks for spotting. Current figure 8 and corresponding Table A1 are older versions of the summary statistics (partially different unit numbering and duplicates in in situ data). The latest version values are available in the figures of appendix B.

Comment: Add colour legend to Figure 10. For figures 15, 16 (and uppermost in 17?) colour legend information is said to be found in Table 2. It might work like that, but does not look very clear to me.

• Reply: Thanks for spotting. The legend information can be also found in Table 2 in this case.