the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
A 30 m annual cropland dataset of China from 1986 to 2021
Shengbiao Wu
Bin Chen
Qihao Weng
Peng Gong
Yuqi Bai
Jun Yang
Bing Xu
Abstract. Accurate, detailed, and up-to-date information on cropland extent is crucial for provisioning food security and environmental sustainability. However, because of the complexity of agricultural landscapes and lack of sufficient training samples, it remains challenging to monitor cropland dynamics at high spatial and temporal resolutions across large geographical extents, especially for places where agricultural land use is changing dramatically. Here we developed a novel cost-effective annual cropland mapping framework that integrated time-series Landsat imagery, automated training sample generation, and machine learning and change detection techniques. We implemented the proposed scheme to a cloud computing platform of Google Earth Engine and generated China’s annual cropland dataset (CACD) at a 30 m spatial resolution for the first time. Results demonstrated that our approach was capable of tracking dynamic cropland changes in different agricultural zones. The pixel-wise F1 scores for annual maps and change maps of CACD were 0.79±0.02 and 0.81, respectively. A further cross-product comparison in terms of accuracy assessment, correlations with statistics, and spatial details indicated the precision and robustness of CACD than other datasets. According to our estimation, from 1986 to 2021, China’s total cropland area expanded by 30,300 km2 (1.79 %), which underwent an increase before 2000 but a general decline between 2000–2015 and a slight recovery afterward. Cropland expansion was concentrated in the northwest while the eastern coastal region experienced substantial cropland loss. In addition, we observed 419,342 km2 (17.57 %) of croplands that were abandoned at least once during the study period. The consistent, high-resolution data of CACD can support progress toward sustainable agricultural use and food production in various research applications. The full archive of CACD is freely available at https://doi.org/10.5281/zenodo.7936885 (Tu et al., 2023a).
- Preprint
(3825 KB) - Metadata XML
-
Supplement
(1041 KB) - BibTeX
- EndNote
Ying Tu et al.
Status: open (until 25 Oct 2023)
-
RC1: 'Comment on essd-2023-190', Chong Liu, 11 Jun 2023
reply
The paper A 30 m annual cropland dataset of China from 1986 to 2021 provides a remarkable attempt at creating national knowledge on the spatial and temporal patterns of cropland in China. In general, it is a well-writen and useful study and I enjoy the reading. The following comments are my suggestions for ensuring its messages are clear ad grounded behinad the results.
1. defination of cropland. Can I say that here you excluded all cash crops, like tea garden, citrus, etc (in addition to ugarcane), all of which are widely distributed in Soutern China. If so , it may be necessary you clearly mentioned this point in your manuscript.
2. intercomparision. I am happy the CACD was well validated with some published land cover products. However, it seems all selected reference datasets are single/multiple epoch maps. How about the agreement level with some cropland dynamic products, e.g. https://glad.umd.edu/dataset/croplands. In this way we can directly know how good or the accuracy of changed cropland, including both cropland expansion and loss.
Citation: https://doi.org/10.5194/essd-2023-190-RC1 -
AC1: 'Reply on RC1', Ying Tu, 26 Sep 2023
reply
We appreciate your precious time and constructive comments, which are greatly helpful in improving our manuscript. We have carefully addressed all raised concerns and revised the manuscript accordingly. Please see our point-by-point responses to your specific comments in the attachment.
-
AC1: 'Reply on RC1', Ying Tu, 26 Sep 2023
reply
-
RC2: 'Comment on essd-2023-190', Anonymous Referee #2, 30 Aug 2023
reply
General comments
Long-term and accurate cropland monitoring is quite important for provisioning food security and environmental sustainability. This study developed an annual cropland dataset in China (CACD) from 1986 to 2021 by using a novel cost-effective annual cropland mapping framework that integrated time-series Landsat imagery. The authors have done a good job in training and validation dataset selection and annual cropland mapping. The accuracy assessment indicates that CACD has relatively high reliability. Comparisons between CACD and other cropland datasets show its improvements spatially. Overall, I think the CACD is a good annual cropland extent dataset with fine resolution. However, I still have some concerns about the methods and results analysis and have been provided in the specific comments.
Specific comments
1. Lines 61-64. You listed two crop type data (i.e., NASS-CDL, European Union 10 m crop type map) and introduced the research gap, but your dataset also does not include the crop type information and making readers a little disappointed. Meanwhile, I can’t agree with “To date, no fine resolution annual cropland dataset of China exists yet”. In your literature review, Yang and Huang (2021) developed the 30 m annual land cover dataset in China (CLUD) from 1990 to 2019. There are no essential differences between cropland in this study and cropland from CLUD, because your dataset also doesn’t include the crop type information.
2. Lines 103. “The aim of this study is to propose a novel paradigm for large-scale fine-resolution cropland dynamics monitoring.” I think the paradigm is not very innovative. A study titled “Forest management in southern China generates short term extensive carbon sequestration” applied a similar framework to analyze the forest dynamics. You two used the same methods: RF-based probability prediction of cropland or forest, and LandTrendr-based segmentation.
3. Lines 116-117. “Cropland in this study is defined as a piece of land of 0.09 ha in minimum (minimum width of 30 m) that is sowed/planted and harvestable at least once within the 12 months after the sowing or planting date.” The definition of cropland in this study differs from that in previous studies. The vegetation indices (e.g., NDVI, EVI) of cropland samplings in the training and validation dataset could reflect the planting or harvest signals. Thus, statistics of vegetation indices variations during the growth period of the samples could improve the reliability rather than depending on visual interpretation only. Additionally, how do you exclude the sugarcane plantation and cassava crop in the training and validation samples? What’s the difference of spectral signals between sugarcane plantation/cassava crop and other crops?
4. Lines 146-147. As you said, “The threshold value was set following recommendations by Ghorbanian et al. (2020)”. But I didn’t find a threshold table to show the difference among the nine agricultural zones. In each subregion, ~800 training samples were used. So, how many cropland and non-cropland samples are there in each subregion?
5. Lines 176-207. I think these two steps are important for the final cropland layer. The authors give two examples (Figure 2 and Figure S2) to illustrate how the LandTrendr algorithm works. I think more examples should be given to prove the robustness of the cropland mapping method. For example, how cropland probabilities and vegetation indices changed when cropland was converted to urban/grassland/forest, and grassland/forest was reclaimed to cropland.
6. Lines 217-218. A spatial-temporal consistency check approach proposed by Li et al. (2015) was applied to refine the annual cropland maps. I don’t think this consistency check algorithm can be used to cropland without any improvements. In Li et al. (2015), there is a very important assumption that “…the transition from urban to other land cover types is not likely and should be avoided… (Section 2.3.2 in Li et al. (2015))”. However, the conversion rule of cropland differs from urban land. More descriptions should be given if there are any improvements to this algorithm.
7. Lines 264-265. Why do western and southeastern coastal areas have relatively low accuracy (F1 score)? Some explanations should be given. Is it because the cropland in southeastern coastal areas more fragmented?
8. Line 316. “Additionally, cropland areas in some inland provinces (such as Guizhou) remained rather stable.” The area of Guizhou province should be rechecked. As I know, Guizhou is the core area of ecological restoration projects of the karst region. Cropland was converted into forest (Yue et al., 2020, Landscape Ecology).
9. Lines 328-330. “In the Ar Horqin Banner of Chifeng city, Inner Mongolia, large-scale croplands were developed for pasture reclamation and cultivation during the past decades”. It should be noted that pasture is a type of grassland rather than crops.
“Similarly, vast agricultural land parcels sprang up in Aksu, Xinjiang for cotton cultivation.” The newly developed dataset doesn’t include crop type information, how do you get this conclusion? Some studies about cotton expansion in Xinjiang Province should be cited to support your conclusion.
10. Lines 336-354. In this part, the authors give much information about cropland abandonment in China. The newly developed shows the cropland loss in the Loess Plateau and Beijing–Tianjin Sand Source Control Project zone (Figure 11). However, there is only a little analysis about the cause of cropland abandonment or cropland loss. For example, cropland loss is mainly driven by the “Grain for Green” ecological project in Shanxi and Inner Mongolia. Cropland abandonment is also affected by factors such as lack of labor and low income (Zhang et al., 2019, Acta Geography Sinica).
11. Figure 3. The cropland and non-cropland samples could be symbolized with different colors
12. Figure 9. The title of the legend is a little weird. “Loss area” should be “Area change” or “Cropland area change”. Additionally, this figure only shows the net change of cropland area. When comparing the total area of cropland gain (increase) and loss during the period, the spatial shift of cropland will be more significant.
Citation: https://doi.org/10.5194/essd-2023-190-RC2 -
AC2: 'Reply on RC2', Ying Tu, 26 Sep 2023
reply
We appreciate your precious time and constructive comments, which are greatly helpful in improving our manuscript. We have carefully addressed all raised concerns and revised the manuscript accordingly. Please see our point-by-point responses to your specific comments in the attachment.
-
AC2: 'Reply on RC2', Ying Tu, 26 Sep 2023
reply
Ying Tu et al.
Data sets
A 30 m annual cropland dataset of China from 1986 to 2021 Ying Tu, Shengbiao Wu, Bin Chen, Qihao Weng, Peng Gong, Yuqi Bai, Jun Yang, Le Yu, Bing Xu https://doi.org/10.5281/zenodo.7936885
Ying Tu et al.
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 916 | 312 | 20 | 1,248 | 56 | 7 | 20 |
- HTML: 916
- PDF: 312
- XML: 20
- Total: 1,248
- Supplement: 56
- BibTeX: 7
- EndNote: 20
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1