Delineating the distribution of mineral and peat soils at the landscape scale in northern boreal regions

Ågren, Anneli M.; Hasselquist, Eliza Maher; Stendahl, Johan; Nilsson, Mats B.; Paul, Siddhartho S.

doi:https://doi.org/10.5194/soil-8-733-2022

Articles | Volume 8, issue 2

https://doi.org/10.5194/soil-8-733-2022

© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/soil-8-733-2022

© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 8, issue 2

Original research article

|

07 Dec 2022

Original research article |

| 07 Dec 2022

Delineating the distribution of mineral and peat soils at the landscape scale in northern boreal regions

Anneli M. Ågren, Eliza Maher Hasselquist, Johan Stendahl, Mats B. Nilsson, and Siddhartho S. Paul

Download

Final revised paper (published on 07 Dec 2022)
Preprint (discussion started on 21 Apr 2022)

Interactive discussion

Status: closed

EC1:
'Comment on egusphere-2022-79', David Dunkerley, 28 Jun 2022

This paper presents a study that sought to revise the mapping of peat soils at national scale across Sweden.
The 'Methods' section reports that the primary data used are detailed elevation data from airborne laser scanning (at 2 m resolution), and a computed soil moisture map previously developed by Ågren et al. The soil moisture map in turn relied on unspecified 'digital terrain indices' (ms. line 143-144), together with "ancillary data on quaternary deposits, soil depth, annual and seasonal runoff etc " (ms. line 145) that were used as input for a machine learning model, to predict soil moisture across Sweden. This section of the Methods presentation seemed inadequate to me. What were the topographic indices? How and at what scale were they derived? How were annual and seasonal runoff quantified, and what was the resolution and quality of these data? Runoff data can surely have been at no finer scale than that of catchment level, in most cases. If so, how can it assist in mapping peat at 2 m resolution? The authors need to explain much more thoroughly the data used and the methods used in the machine learning model. In turn, more commentary was needed on the resolution and quality of the soil moisture maps. What, for instance, is the extent of seasonal variability? Is the parameter calculated perhaps an annual mean or median value?
The predicted soil moisture data were then related to field-mapped peat depths collected from forestry surveys in which pits were excavated, and a regression model was fitted to the data. This is then used to predict peat thicknesses elsewhere across Sweden. However, the relationship between predicted organic layer thickness and measured thickness from the field survey data (Figure 4 in the ms.) shows enormous scatter. The bulk of the data points appear to be for quite thin organic layers (bottom left-hand corner of Fig 4), with relatively few observations > 60 cm (right hand part of Fig 4).
The authors do not actually describe the process of producing their predicted organic layer maps from the soil moisture data, but rather simply jump from Fig 3 to a discussion of the resulting maps. This needs to be corrected.
Given the enormous scatter in Fig 4, the authors at several places say that their thickness maps should not be 'taken literally' (e.g. line 460, line 466) and yet there is no real quantification of the probable magnitude of error at any location. This could have been done by comparing with the field data acquired from pits. The RMSE was reported as 19 cm (line 306) but this is a huge uncertainty given that most of the organic layers appear to be less than 20-30 cm in thickness. Is this level of uncertainty actually acceptable, and are the predicted depths sufficiently reliable for the estimation of carbon stocks, for instance?
Overall, I was left unsure about how much confidence could be placed in the thickness maps generated by the authors. I think that a fuller discussion of actual thicknesses and the likely uncertainty (surely varying with topographic position, and perhaps areal extent of particular organic or peat deposits) in the predictions is required. The authors claim excellent resolution in mapping peat deposits covering just 4 m² (e.g. line 405) - i.e., just a single pixel in data at 2 m resolution. Do such tiny peat deposits actually exist? If so, what accounts for their isolated accumulation? The authors need to comment.
There are minor errors scattered throughout the ms. In particular, I would suggest that as a formal geological Period, 'Quaternary' should be capitalised. This is written 'quaternary' at many places in the ms., and all instances need correcting. The authors are occasionally inconsistent with this, such that Table 2 for instance contains 'Quaternary' as does the heading for Section 2.4, but elsewhere, mostly lower-case letters are used.

Citation: https://doi.org/10.5194/egusphere-2022-79-EC1
- AC1: 'Reply on EC1', Anneli Ågren, 27 Sep 2022
  
  The answers are provided as PDF-files.
  
  Citation: https://doi.org/10.5194/egusphere-2022-79-AC1
RC1:
'Comment on egusphere-2022-79', Anonymous Referee #1, 16 Aug 2022
Ågren et al. map the spatial distribution of peat soils and organic layer thickness in Sweden using an existing soil moisture map and national-level field inventory data. The manuscript is well written and mostly sound, but I was left partly confused when reading the manuscript. I have the following major points

Why did you predict thickness of organic layer based on soil moisture map and not from the original predictor variables that were used to produce the soil moisture map? This seems to be quite odd as there is now double uncertainty in the estimates, as the prediction of soil moisture was already a little uncertain. You should justify your approach better. It would be also interesting to compare different predictor variable sets (e.g. topography variables, satellite imagery etc.) to produce the thickness of organic layer and not just use one existing map.

Based on Figure 4, the fit of the model between soil moisture and organic layer thickness is not very good. This should be discussed in more detail. Actually, there has been a lot of discussion that R2 should not be used for nonlinear regressions. You should justify why you evaluated model performance with R2 and you should also report how you calculated R2.

Precision was lower in your map than in the topographic map. It seems that your mapping approach seems to overpredict peatlands, at least to some extent. This seems to be the case also when visually interpreting the material in Figs. 5 and A1 and looking at the information in Table 3. This should be accounted for and discussed in more detail. You could discuss e.g. why your approach seems to overpredict the extent of peatlands and overestimate thickness of thin organic layers. The overestimation of thickness of thin organic layers is probably due to the selected cubic model. You could potentially also use other (non-linear) regression models and discuss the pros and cons of different models.

Additionally, I have the following more detailed but mostly minor comments:

Abstract:

Please remove the first sentence. It is not necessary.

l17-19: it is not needed to report the results from an existing study. Please rephrase and shorten the sentence

l25: please report also the precision results

l28: “peatlands visible from airplanes” could be written “peatlands that can be visually detected from aerial imagery”

l29: delete “most importantly”

Introduction:

l80-87: the direct quote is unnecessarily long. Do you need to include it?

l113: is the fourth objective necessary to include?

l114: write “study provides a guide to map…”

Methods:

l142-155: this paragraph could be shortened as it describes results from an earlier study, not the methods of this study

Did you account for spatial autocorrelation when e.g., constructing the model and dividing the calibration and validation datasets?

l195: This is difficult to understand. Does it mean that 1:25 000 map covers 1.7% of Sweden and so on?

Why did you include the used accuracy metrics? Kappa has been heavily criticized (see e.g., https://doi.org/10.1016/j.rse.2019.111630). You could also have included F-score.

Section 2.6: How were the field inventory datasets upscaled? Does this simply mean that you calculated national level statistics from the datasets using different methods?

Results

the heading of 3.4 could be changed. Should it be “visual interpretation of peatland maps”?

Discussion and conclusions

l375-376: This is misleading as you used ALS data very indirectly.

l460: you write multiple times that the map should not be taken literally. It is not necessary to mention this multiple times.

The section “The novelty of the developed maps” could be shortened and merged with conclusion section. Some text can also be moved to other parts of discussion.

l509: delete “coarse”, “global mapping” is sufficient.

l504-510: Sentinel-2 has 10 m resolution and it surely can be used for quite detailed planning. There is also other remote sensing than just ALS data that can be used in detailed planning.
Citation: https://doi.org/10.5194/egusphere-2022-79-RC1
- AC2: 'Reply on RC1', Anneli Ågren, 27 Sep 2022
  
  The answers are provided as PDF-files.
  
  Citation: https://doi.org/10.5194/egusphere-2022-79-AC2
RC2:
'Comment on egusphere-2022-79', Anonymous Referee #2, 23 Aug 2022

In this paper the authors present an approach for the identification of peat soils by fitting an empirical relationship between the thickness of the organic layer - measured across Sweden - and a continuous soil moisture map. The article is very interesting and quite well written, but I would prefer to see a bit more detailed description of the procedure followed by the authors.

English language and style are quite fine, only a minor spell check is required.

Most of the references are listed without specifying the number of pages. Please carefully check them, since other mistakes are present. All the references in other languages than English should be clearly indicated.

Other specific comments:

- Line 54: Sewell et al (2020) is not reported in the reference list.

- Line 142: please define the acronym SLU, since the readers could not be familiar with it.

- Line 148: please add references for these statistics.

- Line 387: Arrouays et al (2014) is not reported in the reference list.

- Line 388: Jackson et al (2017) is not reported in the reference list.

- Line 458: Nijp et al (2019) is not reported in the reference list.

I recommend to revise the article. Minor revision is necessary.

Citation: https://doi.org/10.5194/egusphere-2022-79-RC2
- AC3: 'Reply on RC2', Anneli Ågren, 27 Sep 2022
  
  The answers are provided as PDF-files.
  
  Citation: https://doi.org/10.5194/egusphere-2022-79-AC3

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

ED: Publish subject to minor revisions (review by editor) (06 Oct 2022) by David Dunkerley

AR by Anneli Ågren on behalf of the Authors (07 Oct 2022) Author's response Author's tracked changes Manuscript

ED: Publish as is (14 Oct 2022) by David Dunkerley

ED: Publish as is (07 Nov 2022) by Engracia Madejón Rodríguez (Executive editor)

AR by Anneli Ågren on behalf of the Authors (08 Nov 2022)

Short summary

Historically, many peatlands in the boreal region have been drained for timber production. Given the prospects of a drier future due to climate change, wetland restorations are now increasing. Better maps hold the key to insights into restoration targets and land-use management policies, and maps are often the number one decision-support tool. We use an AI-developed soil moisture map based on laser scanning data to illustrate how the mapping of peatlands can be improved across an entire nation.