100-m and 1000-m regular grid observations reveal vegetation-controlled ground surface thermal differentiation in the alpine permafrost of the Headwater Area of the Yellow River

• Two-scale regular grids reveal scale-dependent spatiotemporal heterogeneity with amplified fine-scale TDD and FDD. • Medium FVC (0.5–0.75) optimizes permafrost thermal stability with the lowest MAGST and SO. • Variograms reveal coherent spatial autocorrelation during spring transition but stochastic noise in winter. Ground surface temperature (GST) serves as the upper thermal boundary condition governing the thermal regime of permafrost, yet its fine-scale spatial heterogeneity associated with complex surface characteristics remains a critical uncertainty, largely due to a scarcity of high-density in-situ observations. To bridge this gap, we established a multi-scale systematic monitoring network in June-July 2023 at a representative alpine meadow site in the Headwater Area of the Yellow River, northeastern Qinghai-Xizang Plateau. The network comprises two regular grids with extents of 100 m × 100 m and 1000 m × 1000 m (121 nodes each), capturing hourly GST dynamics. By integrating thermal data with high-resolution (0.1 m) unmanned aerial vehicle (UAV)-derived fractional vegetation cover (FVC), we identified a non-linear vegetation forcing mechanism. A cooling optimum was observed at medium FVC (0.5–0.75), yielding the lowest MAGST (−0.56 ± 0.06 °C and −0.71 ± 0.12 °C for the 100-m and 1000-m grids, respectively) by effectively offsetting summer radiative heating against winter insulation. Conversely, low FVC grid points showed amplified diurnal variability (up to 6.09 °C). Spatial analysis revealed scale-dependent thermal regimes: the fine-scale 100-m grid highlighted localized heat accumulation linked to micro-scale surface heterogeneity, while the 1000-m grid showed seasonal structural instability, where coherent spatial patterns disintegrated during winter. These findings provide critical, scale-dependent constraints for calibrating process-based permafrost models.