Root zone maximum water deficit (S-Rmax) refers to the maximum water consumption of the root zone during drought, which directly influences the partitioning of precipitation between infiltration and runoff. It is a key parameter in land surface hydrological modeling. Since the implementation of the Grain-for-Green Project (GFG) on the Loess Plateau (LP), vegetation restoration has achieved significant success, resulting in the "greening" of LP while simultaneously reducing surface runoff. However, the lack of consideration for the root zone, a key link between terrestrial ecological and hydrological processes, has hindered understanding of ecohydrological mechanisms and limited comprehensive assessments of regional water resource management and ecological engineering outcomes. This study analyzes the spatiotemporal dynamic of S-Rmax on the LP from 1982 to 2018 using multi-source datasets and the Mass Curve Technique. Additionally, we employ a hybrid machine learning-statistical attribution model to quantify the contributions of land use and climate change to the S-Rmax dynamic. The results indicate an average S-Rmax of 85.3 mm across the LP, with significant variations among land use types: natural forest (116.3 mm) > planted forest (104.6 mm) > grassland (87.0 mm) > cropland (78.8 mm). Following the implementation of GFG, S-Rmax increased by 37.7%, with an upward trend observed across all land use types, particularly in changed land type, which experienced the largest increases. The attribution model achieved a coefficient of determination (R-2) of 0.92. The key factors driving S-Rmax variation varied by land use type: in unchanged land type, climate change accounted for 53.8% of the S-Rmax increase, whereas land use change explained 71.3% of the increase in changed land type, with GFG contributing 52.1%. These findings provide a scientific basis for enhancing drought resilience and implementing the "Water-for-Greening" strategy on the LP and similar regions under changing environmental conditions.