| 13 Jan 2026
OasisMap30: A 30 m annual land cover dataset of China's oases from 1987 to 2024
Abstract. High spatio-temporal resolution maps of oasis land cover are valuable for understanding ecological and societal development processes in dryland regions. However, the relatively late development of oasis research, combined with the highly fragmented structure of oases and their frequent land cover transitions, has made it challenging to construct high spatio-temporal resolution land cover datasets for oases. As a result, dedicated data products of this kind are still lacking. Here, we developed a framework for annual 30 m land cover mapping that integrates Landsat satellite imagery, machine learning, a temporal segmentation approach (LandTrendr), and principal component analysis. Using this framework, we produced a 30 m resolution annual land-cover dataset for Chinese oases (OasisMap30) for the period 1987–2024 on the Google Earth Engine (GEE) platform. Accuracy assessment based on more than 6,300 visually interpreted samples demonstrates high accuracy of OasisMap30 (overall accuracy >90 %). In cross-product comparisons based on visually interpreted and third-party test samples, OasisMap30 exhibits considerable advantages in terms of classification accuracy and error reduction. Moreover, in comparison with several 30 m resolution thematic products for impervious surface, cropland, and surface water, we found an impressive consistency between OasisMap30 and these datasets. Using OasisMap30, we investigated the changes in land cover patterns of Chinese oases. The results show that oasis area expanded by 45.87 % (+7.75 Mha) between 1987 and 2024, primarily driven by cropland expansion and grassland restoration. Specifically, 4.04 Mha of desert were restored to grassland, and 3.19 Mha were converted from desert to cropland. In addition, OasisMap30 reveals the expansion of impervious surfaces (0.58 Mha) and surface water (0.35 Mha), as well as conversions among land cover types, such as the conversion of 3.12 Mha of grassland to cropland. Overall, the consistent, high-resolution OasisMap30 data can substantially support studies on the evolution of oasis landscape patterns, socio-ecological responses, and spatial pattern optimization, thereby contributing to the sustainable development of oasis regions. The full archive of OasisMap30 is freely available at https://doi.org/10.6084/m9.figshare.30798032 (Chen et al., 2025).
This manuscript introduces a long-term, high-resolution oasis land-cover dataset, which is valuable for understanding the spatio-temporal dynamics of oasis systems and for related environmental and resource management applications. However, the current manuscript still contains several issues, including technical details and result presentation, that need to be addressed. My specific comments are provided below.
Major comments:
Minor revision:
Line 97 The manuscript states that the tasseled cap transformation was applied using coefficients from Crist et al. (1985). Given that this study uses Landsat 5, 7, and 8 data, the authors should check and clarify whether the cited coefficients are appropriate for all sensors involved.
Line 126 “…a single composite disturbance metric (MAGPCA)”, “PCA” should be a subscript.
Line 129 “…we obtained DURmode and YODmode maps”, “mode” should be a subscript.
Line 136 In Section 2.6, the manuscript describes an evaluation of the dispersion of the detected year-of-disturbance (YOD). However, the results of this analysis (e.g., in figures or tables) are missing. The authors should provide the relevant results to support this analysis.
Line 195 The manuscript reports accuracy assessment results, but it is unclear whether the training and validation samples are spatially well distributed across the study area. Additional details on the spatial distribution of the samples should be provided.
Line 321 Grammatical error: “producing a high spatio-temporal resolution land cover datasets for oases are essential …”.