Study region The middle reaches of the Yellow River (MRYR), China. Study focus Understanding soil moisture (SM) dynamics is vital, current studies lack integrated analyses of multi-drought types and geomorphic-specific vertical responses. This study fused GLDAS, ERA5, and CMA-Land data via Triple Collocation into a high-accuracy M-SM product (1990-2024) and quantified multi-layer spatiotemporal heterogeneity across five geomorphic regions, evaluated the respond of SM to meteorological (SPEI), soil (SSI), and vegetation (TVDI) droughts, and identified dominant drivers by the Boosted Regression Trees. New insights into the response of soil moisture distribution to different droughts and driving factors (1) The M-SM product improved estimation accuracy over single-source products (R= 0.419 similar to 0.668, RMSE= 0.041 similar to 0.095 m & sup3;/m & sup3;). (2) Multi-layer SM showed significant (p < 0.01) wetting trends with southeast-high and northwest-low patterns where 10 similar to 40 cm and 40 similar to 100 cm layers presented most obvious fluctuations. (3) the respond of SM to drought indices showed distinct vertical heterogeneity. The correlations between SSI/SPEI and SM decreased with soil depth, while TVDI exhibited strong correlation with the SM of 10 similar to 40 cm depth. (4) Driving factors of SM showed distinct vertical stratification and regional heterogeneity. Surface layers were driven by meteorological elements, while deep layers were constrained by soil texture. This study elucidated the spatiotemporal variation patterns of SM in complex geomorphological regions and their respond to different drought conditions. The findings hold significant implications for the scientific management and efficient utilization of water and soil resources in semi-arid complex terrain areas.