A 1-km Daily Land Surface Soil Heat Flux Dataset over the Tibetan Plateau (2000–2024) Using a DenseMLP-Based Transfer Learning Framework

Surface soil heat flux (G₀) is critical to surface energy partitioning and permafrost changes, but high-resolution spatiotemporal datasets over the Tibetan Plateau (TP) are still lacking. This study presents a 1-km spatially resolved daily G₀ dataset covering 2000–2024, generated using a DenseMLP-based transfer learning framework integrating multi-source remote sensing, reanalysis, and in-situ measurement data. Evaluation via the leave‑one‑site‑out cross‑validation strategy yields an overall R² of 0.63, a Bias of 0.0 W m⁻², an MAE of 5.0 W m⁻², and an RMSE of 6.6 W m⁻², confirming reliable point-scale accuracy. The dataset exhibits strong spatial heterogeneity on the TP, with high values (>3.5 W m⁻²) in the arid northern and western regions and low values (<1.0 W m⁻²) in the vegetated southeast. It also shows a distinct seasonal cycle in G₀ and a significant regional warming trend (0.05 W m⁻² yr⁻¹) over the 2000–2024 period. This long-term, high-resolution G₀ dataset provides a solid data basis for studying surface energy balance, permafrost dynamics, and climate change on the TP.