The Qinghai-Tibet Plateau (QTP) is sensitive to climate change with extensive permafrost distribution, affecting the vertical transfer of soil moisture from surface to deep layers. However, the understanding of the spatial–temporal variations and transfer patterns of Soil Moisture (SM) from surface to deep layers is not clear. The Mann–Kendall (MK) trend analysis was applied to quantify the spatial–temporal variation trend of SM at different depths on QTP, and the memory analysis and Hurst index were used to evaluate the spatial–temporal distribution pattern of Soil Moisture Memory (SMM). The spatial pattern of four major factors (precipitation, NDVI, land surface temperature, and snow depth) was explored by Random Forest (RF), which dominate SM dynamics from 2000 to 2020, and the potential factors have been discussed. The results indicated that SM trends intensify with depth, and have significant decreases in the southwest with deep Active Layer Thickness (ALT). SMM varies with shorter durations in the southeast for Layer 1 (0 ~ 7cm) and longer in the northwest, while Layer 4 (100 ~ 289cm) exhibits an opposite pattern, where the regions with more persistent memory extend towards the northwest with increasing depth. Precipitation and surface temperature are the key drivers of SM changes in Layer 1, and the vegetation plays a crucial role in the dynamics of Layer 2 (7 ~ 28 cm), Layer 3 (28 ~ 100 cm) and layer 4 (100 ~ 289 cm). Compared with Soil Temperature (ST), NDVI dominated the SM variation. SM in Layer 1 and Layer 2 is positively correlated with vegetation in the southwestern QTP, the areas with positive concentration gradually shifted towards the southeast where meadow and shrub covered in Layer 3 and Layer 4. In contrast to ST, vegetation is more capable of dominating SM variation.
Fan et al. (2026) studied this question.