<p>Arid and semiarid grasslands are an important&#xa0;component of carbon storage for China's grasslands. Their per-unit carbon balance is tightly coupled to precipitation, vegetation cover, and land use, so small environmental or management changes can produce disproportionate shifts. This sensitivity is especially pronounced in the temperate arid–semiarid transitional zone, where climatic gradients are compressed across short distances and overlain by topographic relief and mixed grazing–cropping mosaics. As a result, climate-driven NEP patterns documented at regional scales may not transfer directly to the local scale, motivating a complementary local-scale analysis. Taking Guyang County, Inner Mongolia, as a case, where grassland dominates land use, this study examined the 2020 spatial patterns of carbon sinks and their local driving factors, complementing existing regional-scale work. Net ecosystem productivity (NEP) was estimated with an improved Carnegie–Ames–Stanford Approach (CASA) model and benchmarked against an upscaled eddy-covariance product (Pearson r = 0.797, p &lt; 0.001; residual bias −42.04 g C·m⁻²·a⁻¹). Spatial clustering was then characterized using global and local autocorrelation, and driving factors were identified&#xa0;using a multiscale geographically weighted regression (MGWR) model. County-mean NEP was 101.13 g C·m⁻²·a⁻¹ (total 493.20 Mg C), with a south-high, north-low pattern. Grasslands accounted for 74.51% of high–high clusters and dry croplands for 69.78% of low–low clusters. In Guyang, maximum annual NDVI showed the smallest MGWR bandwidth and the strongest spatial non-stationarity, while climate and soil variables acted as a&#xa0;near-uniform background. Building on these findings, a three-zone management framework is proposed. The analysis is based on a single year and requires multi-year verification.</p>