the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 24 Feb 2023
A new habitat map of the Lena Delta in Arctic Siberia based on field and remote sensing datasets
Abstract. The Lena Delta is the largest river delta in the Arctic (about 30 000 km2) and prone to rapid changes due to climate warming, associated cryosphere loss and ecological shifts. The delta is characterized by ice-rich permafrost landscapes and consists of geologically and geomorphologically diverse terraces covered with tundra vegetation and of active floodplains, featuring approximately 6 500 km of channels and over 30 000 lakes. Because of its broad landscape and habitat diversity the delta is a biodiversity hotspot with high numbers of nesting and breeding migratory birds, fish, caribou and other mammals and was designated a State Nature Reserve in 1995. Characterizing plant composition, above ground biomass and application of field spectroscopy was a major focus of a 2018 expedition to the delta. These field data collections were linked to Sentinel-2 satellite data to upscale local patterns in land cover and associated habitats to the entire delta. Here, we describe multiple field datasets collected in the Lena Delta during summer 2018 including foliage projective cover (Shevtsova et al., 2021a), above ground biomass (Shevtsova et al., 2021b), and hyperspectral field measurements (Runge et al., 2022, https://doi.pangaea.de/10.1594/PANGAEA.945982). We further describe a detailed Sentinel-2 satellite image-based classification of habitat types for the central Lena Delta (Landgraf et al., 2022), an upscaled classification for the entire Lena Delta (Lisovski et al., 2022), as well as a synthesis product for disturbance regimes (Heim and Lisovski, 2023, https://doi.org/10.5281/zenodo.7575691) in the delta that is based on the classification, the described datasets, and field expertise. We present context and detailed methods of these openly available datasets and show how they can improve our understanding of the rapidly changing Arctic tundra system. The new Lena Delta habitat distribution dataset represents a first baseline against which future observations can be compared. With the link between such detailed habitat type classifications and disturbance regimes future upscaling efforts may provide a better understanding of how Arctic lowland landscapes will respond to climate change and how this will impact land surface processes.
- Preprint
(1554 KB) - Metadata XML
-
Supplement
(3249 KB) - BibTeX
- EndNote
Status: closed (peer review stopped)
-
RC1: 'Comment on essd-2023-36', Anonymous Referee #1, 11 Feb 2024
Arctic wetland ecosystems are vulnerable to climate change, but there are still not a few datasets in the region to study the feedback between these ecosystems and climate change. This study provides a few useful datasets about the vegetation in the Lena Delta, which is a unique region. Particularly, the detailed habitat type map will potentially be the most valuable product provided by this study, which, unfortunately, raises me a few concerns about the quality.
The download links for these datasets provide data in tab-delimited text, I’d suggest also provide in other more readable format, such as excel-readable csv, and particularly for maps, in the format of geotiff or other commonly used image format, and a preview in map form (currently, the preview is not working).
Line 155, 162-163. it is confusing here about homogeneous vegetation type. It is homogeneous at what level? It is supposed to only have one “habitat class” as defined in dataset 4 or what? Also, it is stated that in line 155 all 30×30m plots are homogeneous vegetation types, then later in lines 162-163, there are different methods applied to the plot if vegetation cover was homogenous and if vegetation cover is more diverse.
Dataset 1: The vegetation cover was recorded or measured at the center of each 30×30m plot with a ring of 50cm, and then scaled up to the whole plot. How is this done? And how floristic composition played a role in this process. Would the 30×30m plot include more vegetation species than the center 50cm-radius subplot?
Dataset 2: Again, I think more information about the scaling to the 30×30m plot is needed, and why it is reliable.
Dataset 3: Does each 30×30m plot only include one homogeneous land cover? Are these the same plots as datasets 1 & 2?
3.4.1: what is a S-2 based supervised classification?
3.4.2: what is S-205 2A Level 2A image? This section, without pre-information of the sensors, makes it hard to understand the different types/levels of images.
3.4.2: all spectral bands were resampled to 10m pixel resolution bands. Which bands are 10-m resolution, and what spectral bands does the sensor have?
3..4.3: what is the distribution of the 8626 training pixels? Are they scattered in the classification area domain? If they are clustered or formed from polygons with the same class, the efficient number of training pixels will be much reduced. A map of the training/validation samples can help understand the situation.
What is ESUS?
Line 224: A detailed user’s and producer’s accuracy report as well as the overall classification accuracy is needed here to justify a good classification. The validation samples should be “independently” distributed from the training sample: i.e., they should not come from the same plot/polygon, because of the well-known auto-correlation problem.
Line 229: it seems you only have 26+69 training locations. Which is quite a small training sample. As I mentioned, counting the number of pixels in a continuously distributed polygon is misleading.
3.5.2: how large is the area, e.g. at least how many tiles of sentinel-2 images are required to cover the whole study area?
3.5.3: It is good you have 6500 random points now for training the classifier, but still bad that they all come from the central Lena Delta. Some evidence or proof is needed to show that “reflectance” from other areas is like those in the central data when they are the same class, regarding the complex “Same Object, Different Spectrum" and “Different Objects, Same Spectrum” problem. What is the relative size of the central Lena Delta compared to the whole Lena Delta?
Using the training data from just the central Lena Delta to train a classifier to classify a large area is acceptable, but validation the result still using the location from central Lena Delta is not acceptable at all. This is because the accuracy of the classification of a class is dependent on its dominance, which varies by region. The training samples should be evenly scattered out to the whole classification area domain. A map of training/validation points is needed.
Line 294, can you provide more details about how the upscaling is conducted?
Line 299: here you have it named habitat map, but in the data link you call it land use land cover map. I understand they are the same thing but need to be consistent.
Line 306, it is not clear where/what are the hierarchical level 1 in Figure 3a.
Figure 2: Can you label the length of edges of these squares, so we may immediately understand which one is the main 30×30 m plot, and which are smaller-sized subplots? I could not tell which were 2×2 m plots and, which were 0.5×0.5 m plots.
Figure 3: Why do you use surface reflectance for the classification for producing datase4, but then top of atmosphere reflectance for dataset5? It seems to me it is hard to spectrally differentiate the different habitat classes looking at (b), makes me more suspicious about the overall high accuracy of the classification: I do not think the validation scheme is reliable.
Figure 3: what are B and C inlets, they’re not sand probabilities.
Figure 5: is the disturbance map a reclassification of the habitat map?
Citation: https://doi.org/10.5194/essd-2023-36-RC1 -
AC2: 'Reply on RC1', Simeon Lisovski, 16 May 2024
The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2023-36/essd-2023-36-AC2-supplement.pdf
-
AC2: 'Reply on RC1', Simeon Lisovski, 16 May 2024
-
RC2: 'Comment on essd-2023-36', Anonymous Referee #2, 29 Feb 2024
General Comments:
The manuscript provides valuable insights into the ecological changes within the Lena Delta, leveraging satellite imagery and field data to map vegetation and habitat distributions. This work is timely and contributes significantly to our understanding of Arctic ecosystems' responses to climate change. However, several areas require further detail and clarification to enhance the manuscript's impact and utility for future research.
Major Comments:
Data Source and Acquisition Transparency:
Suggestion: Provide detailed information on data sources, including satellite data acquisition dates, sensor types, and ground truth data origins. This detail will help verify the reliability and applicability of the data used.
Methodology on Multi-source Data Fusion:
Suggestion: If the study integrates multiple data sources (e.g., different satellite sensors, ground measurements), describe the fusion methods, including how resolution, coverage time, and accuracy differences were addressed.
Classification and Mapping Algorithm Selection and Validation:Suggestion: Explain the rationale behind choosing specific classification algorithms (e.g., Random Forest, SVM) and how parameters were adjusted for the Lena Delta's unique environment. Also, provide validation details, including data sources for validation, methods, and specific accuracy metrics.
Spatial Interpolation and Scaling Methods:
Suggestion: Detail the spatial interpolation or scaling methods used, their justification, and potential impacts on dataset accuracy and reliability.
Temporal and Spatial Scale:
Temporal-Spatial Coverage and Representativeness:
Suggestion: Discuss whether the dataset's temporal-spatial coverage adequately represents the Lena Delta's seasonal and multi-year variations and its potential impact on ecosystem change analysis.
Choice of Temporal-Spatial Resolution:
Suggestion: Justify the chosen temporal-spatial resolution, including how the balance between data processing capabilities and analysis needs was achieved and the potential impact of different resolutions on result interpretation.
Accuracy and Uncertainty:
Accuracy Assessment:
Suggestion: Include additional accuracy assessment metrics beyond classification accuracy, such as user accuracy, producer's accuracy, and Kappa coefficient, to comprehensively evaluate the dataset's quality.
Uncertainty Analysis:
Suggestion: Conduct a quantitative analysis of uncertainties in the dataset, including those arising from data sources, classification methods, and choice of temporal-spatial resolutions. Discuss the potential impact of these uncertainties on ecosystem analysis and interpretation.
Minor Comments:Language and Expression: Ensure consistency in terminology throughout the manuscript. For instance, clarify the use of "habitat types" vs. "vegetation types" and maintain consistent use throughout.
Figures and Tables: Enhance the readability of figures by adjusting label sizes and including legends directly on figures for clarity. Ensure tables detailing methodologies are clear and abbreviations are defined.
Supplementary Information: Consider adding supplementary material detailing the technical specifications of satellite images, field equipment, and data analysis algorithms.Conclusion:
This study presents important findings on the Lena Delta's ecological dynamics. Addressing the above suggestions will strengthen the manuscript, making it a valuable resource for the scientific community interested in Arctic ecosystems and climate change impacts.
Citation: https://doi.org/10.5194/essd-2023-36-RC2 -
AC1: 'Reply on RC2', Simeon Lisovski, 16 May 2024
The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2023-36/essd-2023-36-AC1-supplement.pdf
-
AC1: 'Reply on RC2', Simeon Lisovski, 16 May 2024
Status: closed (peer review stopped)
-
RC1: 'Comment on essd-2023-36', Anonymous Referee #1, 11 Feb 2024
Arctic wetland ecosystems are vulnerable to climate change, but there are still not a few datasets in the region to study the feedback between these ecosystems and climate change. This study provides a few useful datasets about the vegetation in the Lena Delta, which is a unique region. Particularly, the detailed habitat type map will potentially be the most valuable product provided by this study, which, unfortunately, raises me a few concerns about the quality.
The download links for these datasets provide data in tab-delimited text, I’d suggest also provide in other more readable format, such as excel-readable csv, and particularly for maps, in the format of geotiff or other commonly used image format, and a preview in map form (currently, the preview is not working).
Line 155, 162-163. it is confusing here about homogeneous vegetation type. It is homogeneous at what level? It is supposed to only have one “habitat class” as defined in dataset 4 or what? Also, it is stated that in line 155 all 30×30m plots are homogeneous vegetation types, then later in lines 162-163, there are different methods applied to the plot if vegetation cover was homogenous and if vegetation cover is more diverse.
Dataset 1: The vegetation cover was recorded or measured at the center of each 30×30m plot with a ring of 50cm, and then scaled up to the whole plot. How is this done? And how floristic composition played a role in this process. Would the 30×30m plot include more vegetation species than the center 50cm-radius subplot?
Dataset 2: Again, I think more information about the scaling to the 30×30m plot is needed, and why it is reliable.
Dataset 3: Does each 30×30m plot only include one homogeneous land cover? Are these the same plots as datasets 1 & 2?
3.4.1: what is a S-2 based supervised classification?
3.4.2: what is S-205 2A Level 2A image? This section, without pre-information of the sensors, makes it hard to understand the different types/levels of images.
3.4.2: all spectral bands were resampled to 10m pixel resolution bands. Which bands are 10-m resolution, and what spectral bands does the sensor have?
3..4.3: what is the distribution of the 8626 training pixels? Are they scattered in the classification area domain? If they are clustered or formed from polygons with the same class, the efficient number of training pixels will be much reduced. A map of the training/validation samples can help understand the situation.
What is ESUS?
Line 224: A detailed user’s and producer’s accuracy report as well as the overall classification accuracy is needed here to justify a good classification. The validation samples should be “independently” distributed from the training sample: i.e., they should not come from the same plot/polygon, because of the well-known auto-correlation problem.
Line 229: it seems you only have 26+69 training locations. Which is quite a small training sample. As I mentioned, counting the number of pixels in a continuously distributed polygon is misleading.
3.5.2: how large is the area, e.g. at least how many tiles of sentinel-2 images are required to cover the whole study area?
3.5.3: It is good you have 6500 random points now for training the classifier, but still bad that they all come from the central Lena Delta. Some evidence or proof is needed to show that “reflectance” from other areas is like those in the central data when they are the same class, regarding the complex “Same Object, Different Spectrum" and “Different Objects, Same Spectrum” problem. What is the relative size of the central Lena Delta compared to the whole Lena Delta?
Using the training data from just the central Lena Delta to train a classifier to classify a large area is acceptable, but validation the result still using the location from central Lena Delta is not acceptable at all. This is because the accuracy of the classification of a class is dependent on its dominance, which varies by region. The training samples should be evenly scattered out to the whole classification area domain. A map of training/validation points is needed.
Line 294, can you provide more details about how the upscaling is conducted?
Line 299: here you have it named habitat map, but in the data link you call it land use land cover map. I understand they are the same thing but need to be consistent.
Line 306, it is not clear where/what are the hierarchical level 1 in Figure 3a.
Figure 2: Can you label the length of edges of these squares, so we may immediately understand which one is the main 30×30 m plot, and which are smaller-sized subplots? I could not tell which were 2×2 m plots and, which were 0.5×0.5 m plots.
Figure 3: Why do you use surface reflectance for the classification for producing datase4, but then top of atmosphere reflectance for dataset5? It seems to me it is hard to spectrally differentiate the different habitat classes looking at (b), makes me more suspicious about the overall high accuracy of the classification: I do not think the validation scheme is reliable.
Figure 3: what are B and C inlets, they’re not sand probabilities.
Figure 5: is the disturbance map a reclassification of the habitat map?
Citation: https://doi.org/10.5194/essd-2023-36-RC1 -
AC2: 'Reply on RC1', Simeon Lisovski, 16 May 2024
The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2023-36/essd-2023-36-AC2-supplement.pdf
-
AC2: 'Reply on RC1', Simeon Lisovski, 16 May 2024
-
RC2: 'Comment on essd-2023-36', Anonymous Referee #2, 29 Feb 2024
General Comments:
The manuscript provides valuable insights into the ecological changes within the Lena Delta, leveraging satellite imagery and field data to map vegetation and habitat distributions. This work is timely and contributes significantly to our understanding of Arctic ecosystems' responses to climate change. However, several areas require further detail and clarification to enhance the manuscript's impact and utility for future research.
Major Comments:
Data Source and Acquisition Transparency:
Suggestion: Provide detailed information on data sources, including satellite data acquisition dates, sensor types, and ground truth data origins. This detail will help verify the reliability and applicability of the data used.
Methodology on Multi-source Data Fusion:
Suggestion: If the study integrates multiple data sources (e.g., different satellite sensors, ground measurements), describe the fusion methods, including how resolution, coverage time, and accuracy differences were addressed.
Classification and Mapping Algorithm Selection and Validation:Suggestion: Explain the rationale behind choosing specific classification algorithms (e.g., Random Forest, SVM) and how parameters were adjusted for the Lena Delta's unique environment. Also, provide validation details, including data sources for validation, methods, and specific accuracy metrics.
Spatial Interpolation and Scaling Methods:
Suggestion: Detail the spatial interpolation or scaling methods used, their justification, and potential impacts on dataset accuracy and reliability.
Temporal and Spatial Scale:
Temporal-Spatial Coverage and Representativeness:
Suggestion: Discuss whether the dataset's temporal-spatial coverage adequately represents the Lena Delta's seasonal and multi-year variations and its potential impact on ecosystem change analysis.
Choice of Temporal-Spatial Resolution:
Suggestion: Justify the chosen temporal-spatial resolution, including how the balance between data processing capabilities and analysis needs was achieved and the potential impact of different resolutions on result interpretation.
Accuracy and Uncertainty:
Accuracy Assessment:
Suggestion: Include additional accuracy assessment metrics beyond classification accuracy, such as user accuracy, producer's accuracy, and Kappa coefficient, to comprehensively evaluate the dataset's quality.
Uncertainty Analysis:
Suggestion: Conduct a quantitative analysis of uncertainties in the dataset, including those arising from data sources, classification methods, and choice of temporal-spatial resolutions. Discuss the potential impact of these uncertainties on ecosystem analysis and interpretation.
Minor Comments:Language and Expression: Ensure consistency in terminology throughout the manuscript. For instance, clarify the use of "habitat types" vs. "vegetation types" and maintain consistent use throughout.
Figures and Tables: Enhance the readability of figures by adjusting label sizes and including legends directly on figures for clarity. Ensure tables detailing methodologies are clear and abbreviations are defined.
Supplementary Information: Consider adding supplementary material detailing the technical specifications of satellite images, field equipment, and data analysis algorithms.Conclusion:
This study presents important findings on the Lena Delta's ecological dynamics. Addressing the above suggestions will strengthen the manuscript, making it a valuable resource for the scientific community interested in Arctic ecosystems and climate change impacts.
Citation: https://doi.org/10.5194/essd-2023-36-RC2 -
AC1: 'Reply on RC2', Simeon Lisovski, 16 May 2024
The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2023-36/essd-2023-36-AC1-supplement.pdf
-
AC1: 'Reply on RC2', Simeon Lisovski, 16 May 2024
Data sets
Foliage projective cover of 26 vegetation sites of central Lena Delta from 2018 Iuliia Shevtsova, Nikolay Laschinskiy, Birgit Heim, and Ulrike Herzschuh https://doi.pangaea.de/10.1594/PANGAEA.935875
Total above-ground biomass of 25 vegetation sites of central Lena Delta from 2018 Iuliia Shevtsova, Birgit Heim, Alexandra Runge, Matthias Fuchs, Jan Melchert, and Ulrike Herzschuh https://doi.pangaea.de/10.1594/PANGAEA.935923
Hyperspectral field spectrometry of Arctic vegetation units in the central Lena Delta Alexandra Runge, Matthias Fuchs, Iuliia Shevtsova, Nele Landgraf, Birgit Heim, Ulrike Herzschuh, and Guido Grosse https://doi.org/10.1594/PANGAEA.945982
Sentinel-2 derived central Lena Delta land cover classification Nele Landgraf, Iuliia Shevtsova, Bringfried Pflug, and Birgit Heim https://doi.pangaea.de/10.1594/PANGAEA.945057
Lena Delta Land Cover Classification (2018, Sentinel-2) Simeon Lisovski, Alexandra Runge, Ronald Reagan, Iuliia Shevtsova, and Birgit Heim https://doi.pangaea.de/10.1594/PANGAEA.946407
Lena Delta habitat disturbance regimes Simeon Lisovski and Birgit Heim https://zenodo.org/record/7575691
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 785 | 250 | 60 | 1,095 | 99 | 55 | 57 |
- HTML: 785
- PDF: 250
- XML: 60
- Total: 1,095
- Supplement: 99
- BibTeX: 55
- EndNote: 57
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
