A 10 m resolution land cover map of the Tibetan Plateau with detailed vegetation types

Huang, Xingyi; Yin, Yuwei; Feng, Luwei; Tong, Xiaoye; Zhang, Xiaoxin; Li, Jiangrong; Tian, Feng

doi:https://doi.org/10.5194/essd-2023-327

Preprints

https://doi.org/10.5194/essd-2023-327

Preprints

19 Sep 2023

| 19 Sep 2023

Status: a revised version of this preprint was accepted for the journal ESSD and is expected to appear here in due course.

A 10 m resolution land cover map of the Tibetan Plateau with detailed vegetation types

Xingyi Huang, Yuwei Yin, Luwei Feng, Xiaoye Tong, Xiaoxin Zhang, Jiangrong Li, and Feng Tian

Abstract. The Tibetan Plateau (TP) hosts a variety of vegetation types ranging from broadleaved and needle-leaved forests at the lower altitudes and mesic areas to alpine grassland at the higher altitudes and xeric areas. Accurate and detailed mapping of the vegetation distribution on TP is essential for an improved understanding of climate change effects on terrestrial ecosystems. Yet, existing land cover datasets of TP are either provided at a low spatial resolution or have insufficient vegetation types to characterize certain unique TP ecosystems, such as the alpine scree. Here, we produced a 10 m resolution TP land cover map with 12 vegetation classes and 3 non-vegetation classes for the year 2022 (referred as TP_LC10-2022) by leveraging state-of-the-art remote sensing approaches including the Sentinel-1 and Sentinel-2 imagery, environmental and topographic datasets, and 4 machine learning models using Google Earth Engine platform. Our dataset TP_LC10-2022 achieved an overall classification accuracy of 86.5 % with a Kappa coefficient of 0.854. By comparing with 4 existing global land cover products, TP_LC10-2022 showed significant improvements in terms of reflecting local-scale vertical variations in the southeast TP region. Moreover, we found that alpine scree occupied 13.99 % of the TP region which was ignored in existing land cover datasets, and that shrublands occupied 4.63 % of the TP region characterized by distinct forms of deciduous shrublands and evergreen shrublands largely determined by topography and missed in existing land cover datasets. Our dataset provides a solid foundation for further analyses which need accurate delineation of these unique vegetation types in TP. The TP_LC10-2022 dataset and the sample dataset are freely available at https://doi.org/10.5281/zenodo.8228112 and https://doi.org/10.5281/zenodo.8227942 (Huang et al., 2023a) respectively. Additionally, the classification map can be viewed through https://cold-classifier.users.earthengine.app/view/tplc10-2022.

Received: 12 Aug 2023 – Discussion started: 19 Sep 2023

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.

Download & links

Xingyi Huang, Yuwei Yin, Luwei Feng, Xiaoye Tong, Xiaoxin Zhang, Jiangrong Li, and Feng Tian

Status: closed

CC1:
'Comment on essd-2023-327', Qingyu Li, 01 Jan 2024

This study proposes a 10 m resolution land cover map of the Tibetan Plateau with detailed vegetation types. The experimental design is thorough. However, I have some concerns.
[1] Line 89: Precipitation data is 0.05degree resolution, can the resampled 10m data maintain the quality?
[2] Line 90: What is the spatial resolution of temperature data?
[3] Line 116-117: Why combine bare land and impervious area? Because other land cover products usually separate these two classes.
[4] Line 165: As the vegetation will be affected by seasons, have you considered getting the median composites of Sentinel data for each season, and then combining all seasons as the input?
[5] Line 180: This study is based on pixel-based machine learning classification models. The pixel-based approach tends to produce classification with a salt-pepper effect, did you do any post-classification to remove the noise?
[6] Line 185: Why not use the major voting results of all models as the final results?
[7] Line 245: Have you considered comparing the area in each land cover between your classification and other land cover products?
[8] Authors need to elaborate on the discussion section using more references and describe the implications of your product for the sustainable use of available resources in practice, for policy, and research.

Citation: https://doi.org/10.5194/essd-2023-327-CC1
- AC3: 'Reply on CC1', Xingyi Huang, 05 Apr 2024
  
  The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2023-327/essd-2023-327-AC3-supplement.pdf
  
  Citation: https://doi.org/10.5194/essd-2023-327-AC3
RC1:
'Comment on essd-2023-327', Anonymous Referee #1, 03 Jan 2024

The authors introduced a new classification system and produced a detailed land cover map of the Tibet Plateau (TP) area in 2022, which is significant for climate change studies. The method and results are well-presented. However, there are some questions or issues.
1. Lines 242–245 mentioned that the reason for not using time series is the dense cloud cover in southeastern TP. Could you provide a quantification of the cloud coverage in this region?
2. Lines 131–141. The Landsat NDVI time series from 2013 to 2022 was used to assist in selecting samples. I also noted that the study selected dense samples in the southeastern TP. Could the cloud cover in southeastern TP affect the Landsat time series from 2013 to 2022 and subsequently impact the accuracy of the sample selection?
3. In Fig. 3, the NDVI time series for evergreen needle-leaved forest, evergreen broadleaved forest, and evergreen shrubland look very similar. Can the NDVI time series effectively distinguish between these land cover types?
4. What is the proportion of samples that are directly visually interpreted from Google Earth images, samples using NDVI time series as auxiliary, and samples using only NDVI time series without Google Earth images?
5. How can you eliminate the impact of land cover changes that may have occurred between 2013 and 2022 on the Landsat time series used in the sample selection?
6. According to Table A2, impervious surfaces or built-up areas are considered as bareland in the classification system. However, I noticed that built-up areas in cities such as Xining and Lhasa are incorrectly classified as cultivated vegetation and other land cover types in your product. I also noticed that the barelands in your training samples do not seem to include built-up area samples.
7. Lines 159–161. "Interannual" refers to two or more years, but you have only selected images from one year. Did you mean "Annual"?
8. You used almost all bands from Sentinel-2 with four additional indices. The information provided by some of the bands may be duplicated. For example, the wavelengths of B8 and B8A are close. Is it sufficient to use only one of them?
9. In the comparison with other products, these products are from different years. Due to land cover changes, comparison across years will introduce some error. Using validation samples in 2022 is also unfair to products of other years. These issues need to be discussed.
10. Are there plans to update the product annually or any other future research plans?
11. Table 1, VV and VH are backscatter coefficients, not reflectivities. And it needs to be clarified with the direction of transmission and reception.
12. Typo in table 2, "DNSI" -> "NDSI".

Citation: https://doi.org/10.5194/essd-2023-327-RC1
- AC2: 'Reply on RC1', Xingyi Huang, 05 Apr 2024
  
  The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2023-327/essd-2023-327-AC2-supplement.pdf
  
  Citation: https://doi.org/10.5194/essd-2023-327-AC2
RC2:
'Comment on essd-2023-327', Anonymous Referee #2, 27 Feb 2024

General comments:
This manuscript ‘A 10 m resolution land cover map of the Tibetan Plateau with detailed vegetation types’ produced a 10 m resolution TP land cover map to address the issue of low spatial resolution and incomplete vegetation type coverage in the existing TP land cover dataset. The generated TP_LC10-2022 product will be a valuable data for the study of this region, but the method employed in the manuscript lacks innovation. In addition, there are still many areas that need improvement.
Specific comments:
First, it is recommended to separate the data and methods sections. The current structure is somewhat confusing. It is suggested to merge sections 2.1 to 2.2 under the main heading "2. Study Area and Data." Also, starting from section 2.3 to 2.3.5, it is suggested to be included in Section 3 as "3. Methodology."
Secondly, in section 2.3.1, why did the authors state "The advantages of our classification system are as follows"? Here, the authors should introduce the basis for constructing this classification system, rather than directly discussing its advantages. Moreover, the content of this section seems more like it is introducing the basis for constructing the classification system. Moreover, what exactly is the content "Discriminability in Remote Sensing Imagery" trying to state? Is this part related to the classification system? Furthermore, didn't the authors developed their product using Sentinel data? How come 0.3m of Google Earth imagery was involved here?
Thirdly, why were the training and validation sets configured as 4:1? Typically, they are set to 7:3.
Fourth, the content within Section 2.3.3, from "Interannual remote sensing" to "The median compositing method in GEE was applied to process all bands of Sentinel-1 and Sentinel-2," is suggested to be included in Section 2.2.1. The corresponding preprocessing is suggested to describe in the presentation of the Sentinel-2 data.
Fifth, Table 2 requires an explanation for the choice of features, particularly why all bands of Sentinel-2 are utilized. For example, Band 9 is more commonly used for atmospheric monitoring applications and has a resolution of only 60 meters, so why is it included? Additionally, is it truly beneficial to resample coarse-resolution data (kilometer-scale) like CHIRPS and ERA5 to 10 meters and include them as classification features? Such coarse-resolution data may increase the "mosaic effect" of the classification results. But the most surprising thing is that the authors have clearly indicated that the temporal and phenological features can significantly help to distinguish different categories! However, it is surprising that the authors only used the median composited bands as input features instead of using temporal features. Especially for the distinction between deciduous and evergreen vegetation, is it really possible to do it by median composits alone? Furthermore, as can be seen in Figure 3, the distinction between shrubs and woodlands is quite difficult, even when relying on temporal features. Therefore, I am skeptical about the classification accuracy of these last features.
Finally, does this paper introduce any methodological innovations? Please endeavor to highlight the novel aspects of this study.

Citation: https://doi.org/10.5194/essd-2023-327-RC2
- AC1: 'Reply on RC2', Xingyi Huang, 05 Apr 2024
  
  The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2023-327/essd-2023-327-AC1-supplement.pdf
  
  Citation: https://doi.org/10.5194/essd-2023-327-AC1

Status: closed

CC1:
'Comment on essd-2023-327', Qingyu Li, 01 Jan 2024

This study proposes a 10 m resolution land cover map of the Tibetan Plateau with detailed vegetation types. The experimental design is thorough. However, I have some concerns.
[1] Line 89: Precipitation data is 0.05degree resolution, can the resampled 10m data maintain the quality?
[2] Line 90: What is the spatial resolution of temperature data?
[3] Line 116-117: Why combine bare land and impervious area? Because other land cover products usually separate these two classes.
[4] Line 165: As the vegetation will be affected by seasons, have you considered getting the median composites of Sentinel data for each season, and then combining all seasons as the input?
[5] Line 180: This study is based on pixel-based machine learning classification models. The pixel-based approach tends to produce classification with a salt-pepper effect, did you do any post-classification to remove the noise?
[6] Line 185: Why not use the major voting results of all models as the final results?
[7] Line 245: Have you considered comparing the area in each land cover between your classification and other land cover products?
[8] Authors need to elaborate on the discussion section using more references and describe the implications of your product for the sustainable use of available resources in practice, for policy, and research.

Citation: https://doi.org/10.5194/essd-2023-327-CC1
- AC3: 'Reply on CC1', Xingyi Huang, 05 Apr 2024
  
  The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2023-327/essd-2023-327-AC3-supplement.pdf
  
  Citation: https://doi.org/10.5194/essd-2023-327-AC3
RC1:
'Comment on essd-2023-327', Anonymous Referee #1, 03 Jan 2024

The authors introduced a new classification system and produced a detailed land cover map of the Tibet Plateau (TP) area in 2022, which is significant for climate change studies. The method and results are well-presented. However, there are some questions or issues.
1. Lines 242–245 mentioned that the reason for not using time series is the dense cloud cover in southeastern TP. Could you provide a quantification of the cloud coverage in this region?
2. Lines 131–141. The Landsat NDVI time series from 2013 to 2022 was used to assist in selecting samples. I also noted that the study selected dense samples in the southeastern TP. Could the cloud cover in southeastern TP affect the Landsat time series from 2013 to 2022 and subsequently impact the accuracy of the sample selection?
3. In Fig. 3, the NDVI time series for evergreen needle-leaved forest, evergreen broadleaved forest, and evergreen shrubland look very similar. Can the NDVI time series effectively distinguish between these land cover types?
4. What is the proportion of samples that are directly visually interpreted from Google Earth images, samples using NDVI time series as auxiliary, and samples using only NDVI time series without Google Earth images?
5. How can you eliminate the impact of land cover changes that may have occurred between 2013 and 2022 on the Landsat time series used in the sample selection?
6. According to Table A2, impervious surfaces or built-up areas are considered as bareland in the classification system. However, I noticed that built-up areas in cities such as Xining and Lhasa are incorrectly classified as cultivated vegetation and other land cover types in your product. I also noticed that the barelands in your training samples do not seem to include built-up area samples.
7. Lines 159–161. "Interannual" refers to two or more years, but you have only selected images from one year. Did you mean "Annual"?
8. You used almost all bands from Sentinel-2 with four additional indices. The information provided by some of the bands may be duplicated. For example, the wavelengths of B8 and B8A are close. Is it sufficient to use only one of them?
9. In the comparison with other products, these products are from different years. Due to land cover changes, comparison across years will introduce some error. Using validation samples in 2022 is also unfair to products of other years. These issues need to be discussed.
10. Are there plans to update the product annually or any other future research plans?
11. Table 1, VV and VH are backscatter coefficients, not reflectivities. And it needs to be clarified with the direction of transmission and reception.
12. Typo in table 2, "DNSI" -> "NDSI".

Citation: https://doi.org/10.5194/essd-2023-327-RC1
- AC2: 'Reply on RC1', Xingyi Huang, 05 Apr 2024
  
  The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2023-327/essd-2023-327-AC2-supplement.pdf
  
  Citation: https://doi.org/10.5194/essd-2023-327-AC2
RC2:
'Comment on essd-2023-327', Anonymous Referee #2, 27 Feb 2024

General comments:
This manuscript ‘A 10 m resolution land cover map of the Tibetan Plateau with detailed vegetation types’ produced a 10 m resolution TP land cover map to address the issue of low spatial resolution and incomplete vegetation type coverage in the existing TP land cover dataset. The generated TP_LC10-2022 product will be a valuable data for the study of this region, but the method employed in the manuscript lacks innovation. In addition, there are still many areas that need improvement.
Specific comments:
First, it is recommended to separate the data and methods sections. The current structure is somewhat confusing. It is suggested to merge sections 2.1 to 2.2 under the main heading "2. Study Area and Data." Also, starting from section 2.3 to 2.3.5, it is suggested to be included in Section 3 as "3. Methodology."
Secondly, in section 2.3.1, why did the authors state "The advantages of our classification system are as follows"? Here, the authors should introduce the basis for constructing this classification system, rather than directly discussing its advantages. Moreover, the content of this section seems more like it is introducing the basis for constructing the classification system. Moreover, what exactly is the content "Discriminability in Remote Sensing Imagery" trying to state? Is this part related to the classification system? Furthermore, didn't the authors developed their product using Sentinel data? How come 0.3m of Google Earth imagery was involved here?
Thirdly, why were the training and validation sets configured as 4:1? Typically, they are set to 7:3.
Fourth, the content within Section 2.3.3, from "Interannual remote sensing" to "The median compositing method in GEE was applied to process all bands of Sentinel-1 and Sentinel-2," is suggested to be included in Section 2.2.1. The corresponding preprocessing is suggested to describe in the presentation of the Sentinel-2 data.
Fifth, Table 2 requires an explanation for the choice of features, particularly why all bands of Sentinel-2 are utilized. For example, Band 9 is more commonly used for atmospheric monitoring applications and has a resolution of only 60 meters, so why is it included? Additionally, is it truly beneficial to resample coarse-resolution data (kilometer-scale) like CHIRPS and ERA5 to 10 meters and include them as classification features? Such coarse-resolution data may increase the "mosaic effect" of the classification results. But the most surprising thing is that the authors have clearly indicated that the temporal and phenological features can significantly help to distinguish different categories! However, it is surprising that the authors only used the median composited bands as input features instead of using temporal features. Especially for the distinction between deciduous and evergreen vegetation, is it really possible to do it by median composits alone? Furthermore, as can be seen in Figure 3, the distinction between shrubs and woodlands is quite difficult, even when relying on temporal features. Therefore, I am skeptical about the classification accuracy of these last features.
Finally, does this paper introduce any methodological innovations? Please endeavor to highlight the novel aspects of this study.

Citation: https://doi.org/10.5194/essd-2023-327-RC2
- AC1: 'Reply on RC2', Xingyi Huang, 05 Apr 2024
  
  The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2023-327/essd-2023-327-AC1-supplement.pdf
  
  Citation: https://doi.org/10.5194/essd-2023-327-AC1

Xingyi Huang, Yuwei Yin, Luwei Feng, Xiaoye Tong, Xiaoxin Zhang, Jiangrong Li, and Feng Tian

Data sets

A 10 m resolution land cover map of the Tibetan Plateau with detailed vegetation types Xingyi Huang, Yuwei Yin, Luwei Feng, Xiaoye Tong, Xiaoxin Zhang, Jiangrong Li, Feng Tian https://doi.org/10.5281/zenodo.8228112

A dataset of land cover samples over the Tibetan Plateau Xingyi Huang, Yuwei Yin, Luwei Feng, Xiaoye Tong, Xiaoxin Zhang, Jiangrong Li, Feng Tian https://doi.org/10.5281/zenodo.8227942

Xingyi Huang, Yuwei Yin, Luwei Feng, Xiaoye Tong, Xiaoxin Zhang, Jiangrong Li, and Feng Tian

Viewed

Total article views: 1,210 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
872	293	45	1,210	37	50

HTML: 872
PDF: 293
XML: 45
Total: 1,210
BibTeX: 37
EndNote: 50

Views and downloads (calculated since 19 Sep 2023)

Month	HTML	PDF	XML	Total
Sep 2023	239	67	3	309
Oct 2023	147	52	4	203
Nov 2023	50	31	1	82
Dec 2023	57	16	4	77
Jan 2024	63	10	1	74
Feb 2024	58	17	7	82
Mar 2024	64	36	4	104
Apr 2024	69	17	10	96
May 2024	39	19	7	65
Jun 2024	62	21	2	85
Jul 2024	24	7	2	33

Cumulative views and downloads (calculated since 19 Sep 2023)

Month	HTML	PDF	XML	Total
Sep 2023	239	67	3	309
Oct 2023	147	52	4	203
Nov 2023	50	31	1	82
Dec 2023	57	16	4	77
Jan 2024	63	10	1	74
Feb 2024	58	17	7	82
Mar 2024	64	36	4	104
Apr 2024	69	17	10	96
May 2024	39	19	7	65
Jun 2024	62	21	2	85
Jul 2024	24	7	2	33

Viewed (geographical distribution)

Total article views: 1,169 (including HTML, PDF, and XML) Thereof 1,169 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 16 Jul 2024

Short summary

The Tibetan Plateau, with its diverse vegetation from forests to alpine grasslands, plays a key role in understanding climate change impacts. Existing maps lack detail or miss unique ecosystems. Our research, using advanced satellite technology and machine learning, produced the TP_LC10-2022 map. Comparisons with other maps revealed TP_LC10-2022's excellence in capturing local variations. Our map is significant for in-depth ecological studies.


Total:	0
HTML:	0
PDF:	0
XML:	0