A 1-km Daily Land Surface Soil Heat Flux Dataset over the Tibetan Plateau (2000–2020)

This dataset provides daily soil heat flux (G₀) data across the Tibetan Plateau (TP) for the period 2000–2020, with a spatial resolution of 1 km. It addresses the existing gap in long-term, high-spatial-resolution, and spatially continuous G₀ observations over the TP, which is of great significance for deepening the understanding of surface energy partitioning, permafrost evolution, and land-atmosphere interaction processes in high-altitude cold regions like the Tibetan Plateau. The dataset was constructed based on a DenseMLP transfer learning framework, integrating diverse multi-source datasets to guarantee its accuracy and spatial representativeness. The input datasets include land surface temperature from the TRIMS dataset, vegetation indices (NDVI, EVI) derived from MOD13Q1, meteorological forcings (surface soil heat flux, soil moisture, net shortwave radiation, net longwave radiation) extracted from GLDAS, static soil properties (bulk density, coarse fragments, sand/silt/clay fractions, soil organic carbon) from SoilGrids, and topographic elevation from MERIT DEM. Site-scale validation was conducted to evaluate the performance of this dataset, and the results confirm its robustness. Specifically, on the training set, the dataset achieves an overall coefficient of determination (R²) of 0.69, a Bias of 0.10 W m⁻², a mean absolute error (MAE) of 3.68 W m⁻², and a root mean square error (RMSE) of 4.95 W m⁻². On the independent test set, it yields an R² of 0.71, a Bias of 0.70 W m⁻², an MAE of 5.09 W m⁻², and an RMSE of 6.53 W m⁻². These validation metrics indicate good consistency between the dataset and in-situ observations, with no significant overfitting, thereby verifying the reliability of the dataset for estimating daily 1-km G₀ over the TP. This high-resolution G₀ dataset offers new insights and reliable data support for investigating land–atmosphere interactions and eco-environmental responses to global climate change in high-altitude regions, particularly the TP. Furthermore, it has considerable potential for wide-ranging applications in permafrost dynamic monitoring and land surface model validation.