The ecological and environmental safety of the Kherlen River Basin is directly related to the sustainable development of both China and Mongolia. Scientific delineation of non-point source pollution control units is crucial for precise implementation of water environment policies and efficient management in the basin. However, currently, there is a lack of effective zoning data to guide specific measures in pollution control in this region. Traditional methods of dividing pollution control units struggle to accurately reflect the differences in grassland non-point source pollution, thereby affecting management effectiveness to some extent. Grassland non-point source pollution is influenced by multiple factors, exhibiting both attribute repetition and spatial continuity. To capture these characteristics more accurately, a clustering method that balances attribute repetition and spatial continuity is required. In this study, focusing on the Kherlen River Basin and targeting the influencing factors of grassland non-point source pollution, we comprehensively considered key continuous data such as annual average precipitation, temperature, digital elevation, grassland carrying capacity, and soil total nitrogen and phosphorus content. Utilizing the Spatial Toeplitz Inverse Covariance Clustering (STICC) method, which effectively handles attribute dependencies and spatial consistency strategies, we conducted clustering analysis and constructed a 2022 dataset for non-point source pollution control zoning in the Kherlen River Basin. To validate the accuracy of this dataset, we compared the zoning effects using the DUNN clustering accuracy evaluation index with other traditional zoning results. The results showed that the STICC method outperforms methods like K-Means, Spectral K-Means, GMM, and Repeated Bisection in clustering accuracy. It can more effectively identify heterogeneous pollution areas, significantly enhancing the precision of management. Additionally, this study preserved the original continuity of the data, resulting in a more accurate depiction of pollution characteristics. Compared to traditional methods, the zoning data provided in this study improves detail presentation by more than 50%. This dataset not only offers strong support for in-depth studies on non-point source pollution characteristics in the Kherlen River Basin but also provides a solid data foundation for related control decisions.