Extended Data Fig. 2: Schematic of the space-for-time approach to derive the sensitivities of various climatic factors (for example, VPD) to the deepening ALT.

This method assumes that the adjacent pixels share the same background climate with the central pixel within the spatial moving window (set as 9×9 pixels in our study). It can therefore detect the one-way signal representing the climatic feedback from the deepening ALT. a-b, Adjacent pixels were selected based on criteria that can exclude the confounding effects of vegetation dynamics (LAI and FVC), land-cover changes (main land-cover type, LC; fraction of main land-cover type, LC_fraction), topography (DEM), and permafrost type. c, Theil-Sen slope estimator was applied to estimate sensitivity by pooling all selected adjacent pixels and the central pixel itself together. This method was also used to estimate the sensitivities of vegetation growth (FVC and LAI) and below-ground fuel dryness (FFMC, DMC, and DC) to ALT. The constraints of vegetation dynamics on the selection of adjacent pixels were not considered when calculating the sensitivity of FVC and LAI to ALT. The shaded area in c indicates the 95% confidence interval of fitting line. Maps created in ArcGIS (v.10.8) using a base map derived from the IPCC-WG1 Atlas GitHub repository⁸⁰.