This dataset is a spatial distribution product of the volumetric moisture content of the active layer in the permafrost region of the Qinghai Tibet Plateau. It is produced to meet the research needs of permafrost hydrothermal processes, alpine hydrology, and ecological environment. It is the first high-resolution active layer moisture dataset in the region to cover the entire permafrost region, filling the data gap in continuous spatial mapping of active layer moisture at all depths on the Qinghai Tibet Plateau.
The data is based on 342 measured points in the permafrost region of the Qinghai Tibet Plateau from 2009 to 2024, including profile data measured by the ring knife method and ground penetrating radar (GPR) inversion data; Collaborative multi-source remote sensing factors (Sentinel-1 inversion NDVI、NDWI、 Modeling and generation of public goods such as surface temperature LST, SRTM 90m terrain factor, and China Land Surface Simulation Soil Attribute Dataset (CSDLv2).
Four integrated machine learning models (random forest, extreme random tree, XGBoost, CatBoost) were used for fusion, and SHAP recursive feature elimination was used to optimize feature combination. Five fold cross validation was used to ensure the model's generalization ability, and a bias aware multi model fusion strategy was adopted to correct system biases with low overestimation and high underestimation. Finally, a continuous spatial distribution of active layer moisture was obtained for the entire domain.
The data is a single band GeoTIFF grid, representing the annual average volume water content of the active layer during the peak melting period (September to October), in m ³ · m ⁻ ³. The spatial range is the permafrost region of the Qinghai Tibet Plateau, with a geographic coordinate system of GCSWGS_1984 and a spatial resolution of 90m. The time attribute is a comprehensive static cross-section for many years, without a time period; The file naming convention is clear and easy to batch call and identify.
The data quality has been cross validated, with a determination coefficient R ² of 0.59-0.61, root mean square error RMSE ≈ 0.08 m ³ · m ⁻ ³, and average absolute error MAE ≈ 0.06 m ³ · m ⁻ ³. Quality control covers the entire process of sample screening, outlier removal, feature selection, model fusion, and deviation correction, and the spatial pattern is reliable.
The advantage of this dataset is its 90m high resolution, full coverage of frozen soil, and focus on the depth of the entire active layer. Compared to traditional surface soil moisture products, it is more suitable for simulating frozen soil water and heat; It can be applied to monitoring permafrost changes, simulating hydrological processes in high-altitude regions, ecological hydrological assessment, climate response research, and land surface model data assimilation.
| # | number | name | type |
| 1 | XDB0950000 | Temporal and spatial changes of frozen soil and underground ice and improvement of engineering service performance | Strategy Priority Research Program (Category B) of Chinese Academy of Sciences |
| 2 | 42525105 | geography of permafrost | National Natural Science Foundation of China Youth Science Fund Project |
| 3 | 2025NCDC010 | AI-ready data preparation and spatial distribution simulation of active layer moisture in the Qinghai-Tibet Plateau based on GPR-remote sensing-physical and chemical multi-source fusion and machine learning | NCDC Open Fund Project |
This work is licensed under
CC BY 4.0 (Creative Commons Attribution 4.0 International License).
CSTR:11738.11.NCDC.PERMAFROST.DB7312.2026
10.12072/ncdc.permafrost.db7312.2026
