Precipitation use efficiency (PUE) and soil water use efficiency (SWUE) are key indicators of the carbon-water cycle in terrestrial ecosystems. Yet in water-limited regions, the spatial patterns and driving mechanisms of this cycle, shaped by precipitation and soil moisture, are still poorly understood. This study used an improved twoleaf model with modified environmental stress factor expressions to simulate gross primary production (GPP) on the Loess Plateau from 2001 to 2023, then calculated PUE and SWUE by integrating precipitation and soil moisture data. To investigate the spatial dynamics and driving mechanisms of these indicators, we used linear regression, Mann-Kendall tests, XGBoost, and residual analysis. The results show that the improved two-leaf model's simulations closely match observations, reducing root mean square error by 21.8%. Both PUE and SWUE have exhibited an increasing trend, with 64.34% of the area showing simultaneous increases, and the multi-year average exhibiting a gradient shift from northwest to southeast. Compared with other vegetation types, cropland shows the highest annual mean PUE and SWUE, with the greatest rate of change occurring in cropland reclamation areas. PUE and SWUE are influenced by multiple factors. Precipitation and the aridity index carry relatively high weights in controlling PUE, whereas soil moisture is the primary environmental determinant of SWUE, and its influence increased markedly from 2001 to 2023. Moreover, human activities have a greater impact on both PUE and SWUE than climate change. These findings provide scientific guidance for vegetation restoration and water-resource management in arid and semi-arid regions.