the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 10 Aug 2026
Three decades of grassland canopy cover change in China (1990–2023) revealed by high-resolution mapping using extensive drone and Landsat data
Abstract. Long-term, high-resolution canopy cover data are essential for understanding grassland ecosystem dynamics and informing sustainable management. However, existing products are largely limited to coarse spatial resolutions, constraining their utility for high-precision, large-scale analyses. In this study, we collected over 16,000 drone image tiles (30 m × 30 m) from 2,144 sites across China and developed a machine learning model to produce a spatially seamless, 30 m annual dataset of national grassland canopy cover from 1990 to 2023 by integrating drone and Landsat-series imagery. The model achieves high predictive accuracy (R2 = 0.73, RMSE = 18.4 %) and robust temporal transferability (R2 = 0.68, RMSE = 20.6 %). Comparisons with existing large-scale products demonstrated significantly improved accuracy and reduced residual artifacts, underscoring the robustness of our approach across diverse grassland types and time periods. Spatiotemporal analysis indicated a multi-decadal mean canopy cover of 43.80 ± 18.69 % across China’s grasslands. Over the 34-year period, 41.76 % of grasslands exhibited significant increases, 57.16 % showed nonsignificant change, and 1.08 % experienced significant declines. Climatic factors—including drought, precipitation, and temperature—emerged as the dominant drivers of canopy cover dynamics at the national scale, although their effects exhibited pronounced spatial heterogeneity. In contrast, anthropogenic pressures played a secondary role overall but could override climatic influences at local scales. Collectively, these findings, together with the long-term, high-resolution canopy cover dataset developed in this study, provide an essential basis for advancing the understanding of grassland ecosystem dynamics and for supporting evidence-based conservation and sustainable management strategies, particularly under intensifying climate change and increasing frequency of extreme events. The national grassland canopy cover dataset generated in this study is archived on Zenodo and can be freely downloaded from https://doi.org/10.5281/zenodo.20301123 (Jiang et al., 2026).
- Preprint
(3941 KB) - Metadata XML
-
Supplement
(1895 KB) - BibTeX
- EndNote
Status: open (until 21 Oct 2026)
- RC1: 'Comment on essd-2026-628', Anonymous Referee #1, 09 Oct 2026 reply
-
RC2: 'Comment on essd-2026-628', Anonymous Referee #2, 10 Oct 2026
reply
This manuscript combines nationwide UAV surveys with Landsat imagery to produce a useful dataset of grassland canopy cover across China for 1990–2023. However, it is unclear whether the reported accuracy applies to the final annual product, and the historical validation does not directly test changes in cover at the same locations. I recommend major revision, focusing on issues that affect the use of the dataset and the interpretation of long-term changes.
General comments
- Please clarify which steps use Landsat scenes closest to the reference dates and which use annual composites. If validation uses the former, does the accuracy reported in Fig. 4 apply to the annual peak-cover product?
- Each site was divided into nine image tiles. Were the training and validation sets separated by site?
- The historical validation mainly tests the accuracy of canopy cover values. I suggest using repeat observations from the same sites to assess whether the model also captures observed changes in cover.
- The standard deviation across models in Fig. 13 does not account for all sources of error. Please explain how it differs from the validation RMSE, so that users do not underestimate the product’s uncertainty.
- How were the reference values matched to the pixel size of the 250 m products? If 30 m reference samples were used directly, differences in spatial scale could affect the comparison.
- The driver model explains only 36% of the variation, and human activity is represented by just two indicators. The conclusion that human activity plays a relatively minor role needs to be qualified.
Specific comments
- Lines 137–139: Was the accuracy of the existing MLP model checked on the imagery used in this study? Errors in the reference cover estimates would also affect the RF model.
- Lines 219–221: Which variable was used to select observations for the annual maximum-value composite? Were all spectral bands taken from the same acquisition date?
- Lines 222–223: Please report the range of time intervals between the reference imagery and Landsat acquisitions, and discuss whether these gaps introduce appreciable phenological differences.
- Fig. 4: Please report validation accuracy by major grassland type. The pooled results are insufficient to assess reliability for types with few reference samples, particularly alpine steppe.
- Fig. 4: How large are the overestimation at low cover and underestimation at high cover? These biases could affect estimates of the magnitude of degradation and recovery.
- Lines 275–276: The reported area fractions of cover gains and losses apply only to pixels classified as stable grassland. The abstract and conclusions should make this scope clear.
- Line 201: Driver trends were calculated through 2020, whereas canopy cover trends extend through 2023. Why were different periods used?
- Fig. 8: Was R² = 0.36 calculated on the training data or an independent validation set? Section 2.5 does not describe how the driver model was validated.
- Lines 473–475: Fig. 13 assesses uncertainty only for 2021. It does not, on its own, support the explanation concerning uncertainty introduced by early Landsat imagery.
Citation: https://doi.org/10.5194/essd-2026-628-RC2 -
RC3: 'Comment on essd-2026-628', Hongliang Ma, 10 Oct 2026
reply
This paper developed a high-resolution grassland canopy cover product for China from 1990 to 2023 by combining extensive UAV observations with Landsat imagery. The dataset is potentially valuable for understanding long-term grassland dynamics, and the manuscript is generally well organized and clearly presented. The presented topic is also interesting and falls within the scope of ESSD. Before recommending it for publication, I have several moderate comments on its current form and hope it will be helpful for the improvements:
Abstract:
P2L53: canopy cover-> grassland canopy cover
L54: Please clarify the specific resolution.
L59: when stating this, what's the baseline product or ground truth?
L63: Please improve the expression and remove the extra dash, which may be the sign of AI-generated sentences.
it is not common to list drought with precipitation and temperature...and which temperature, please clarify.
Introduction:
P3L89: due to mixed-pixel issue at the level of 500m?
Methods:
L189: please clarify how do you consider the uncertainties due spatial-temporal mismatching issues regarding these multi-source data?
L243-259:
How reliable is the temporal transferability of the model, given that it was trained using observations from 2021–2022 but applied to 1990–2023, with no independent validation before 2002?
L244: how about the spatial transferability? See the examples for spatial-independent validation in one previous paper (‘Surface soil moisture from combined active and passive microwave observations: Integrating ASCAT and SMAP observations based on machine learning approaches’)
Discussion:
Please enhance the comparisons with existing similar efforts, particularly for similar products, algorithms, accuracy and improvements. Here I may suggest adding one comprehensive table to present this point, which should be very helpful (see Section 5.3.2. in Ma et al. 2024).
Citation: https://doi.org/10.5194/essd-2026-628-RC3
Data sets
Annual 30 m grassland canopy cover dataset for China from 1990 to 2023 Xuanhao Jiang, Tianyu Hu, Zhenying Huang, Zhiyao Tang, Leiyi Chen, Wenli Chen, Xiaoli Cheng, Yunxiang Cheng, Yuejun Fan, Ruiru Gao, Tiangang Gao, Yanming Gong, Yanpei Guo, Xiaokang Hu, Chengjun Ji, Liping Li, Shijin Li, Guofang Liu, Wanyin Luo, Hongyuan Ma, Wenhong Ma, Xiujuan Qiao, Chuangye Song, Guoyan Wang, Lei Wang, Qinggang Wang, Yonghui Wang, Zhong Wang, Xuejun Yang, Yan Yang, Yongchuan Yang, Xuehua Ye, Chunying Yin, Feihai Yu, Pujin Zhang, Shuren Zhang, Chengyang Zheng, Luhong Zhou, Yanjun Su https://doi.org/10.5281/zenodo.20301123
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 265 | 103 | 139 | 507 | 75 | 168 | 115 |
- HTML: 265
- PDF: 103
- XML: 139
- Total: 507
- Supplement: 75
- BibTeX: 168
- EndNote: 115
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The manuscript presents a spatially seamless, annual 30-m dataset of grassland canopy cover across China from 1990 to 2023, developed by integrating UAV observations with Landsat imagery. The topic is well suited to ESSD, and the dataset is potentially valuable. However, several concerns regarding the machine-learning framework and result validation need to be addressed before the manuscript can be considered for publication.
General comments:
Specific comments:
1) P8L177: To reduce inter-sensor variability, the authors harmonized TM/ETM+ data using OLI as the reference. Figure S1 shows the images before and after calibration. It would be helpful to additionally provide time series of representative pixels before and after calibration to demonstrate temporal consistency.
2) P10L216: The five vegetation indices are likely to be correlated, and similar correlations may exist among the topographic variables. Please provide a correlation analysis among these predictor variables and discuss potential multicollinearity.
3) Fig. 4: The model tends to overestimate low-to-moderate canopy cover and underestimate high canopy cover (>80%). Please clarify whether this systematic bias affects the magnitude and reliability of the estimated long-term trends.
4) Fig. 6: Stratifying canopy cover by province may have limited ecological relevance. Consider grouping the results by climate zones, elevation ranges, or other environmental conditions. A similar issue applies to Fig. 7.
5) P18L345: The five climatic and anthropogenic variables explain only 36% of the variation in grassland canopy cover dynamics. Please discuss the unexplained variation and associated uncertainty, and avoid overinterpreting the SHAP results as causal relationships.
6) Figs. 11 and 12: Fig. 12 shows an apparently similar spatial pattern between the proposed product and the MultiVI FVC product, whereas their R2 in Fig. 11 is relatively low and even lower than those with MODIS VCF and China FVC. Please explain this apparent discrepancy.