<p>Understanding the driving mechanisms of carbon storage at the county level is critical to achieving China’s carbon neutrality goals. While driving factors of carbon storage have been extensively studied, their nonlinear responses and threshold effects to natural and anthropogenic factors remain inadequately clarified. Focusing on the Loess Plateau (LP) as a representative ecologically fragile region, this study established an integrated framework for ‘nonlinear effect analysis-threshold identification-potential zoning’ to reveal the county-level nonlinear responses and threshold effects of carbon storage. This framework was developed by integrating the InVEST, Random Forest, Shapley Additive Explanations, and the Sigmoid function. The results indicated that the carbon storage in the LP continuously increased from 1985 to 2019, reaching 44.54 × 10⁸ t in 2019; NDVI and slope consistently dominated as the primary drivers, while the importance of the proportion of construction land (PCL) rose from 2.97% to 9.65%; County-level carbon storage demonstrated nonlinear responses to multiple drivers, accompanied by temporally varying threshold effects. In 2019, the effect on carbon storage shifted from negative to positive when NDVI &gt; 0.60, slope &gt; 11.03°, while transitioning from positive to negative when DEM &gt; 1747&#xa0;m, PCL &gt; 15.11%, and Nighttime light &gt; 1656. Based on the carbon storage potential index, 37.73% of the counties were classified as high-potential zone, predominantly located in the southeast. These findings demonstrate that effective carbon storage enhancement management in ecologically fragile areas requires implementing county-level precision management strategies guided by the threshold characteristics of key driving factors and the system’s potential status.</p>