Objectives  Buffer analysis is a common tool of spatial analysis, which deals with the problem of proximity. Due to numerous and complex operations in the algorithm, the computational efficiency needs to be optimized.

  Methods  To process large scale point features, a graph-based representation model is proposed, which establishes the spatial computational domain for data and analysis, and develops a well-balanced task-partitioning method by partitioning the graph. First, the proposed model defines processing functions of point features and their spatial relationships from the perspectives of graph nodes and graph edges, and provides a logic description for buffer zone generation around point features. Second, the computational weights of graph nodes and graph edges are obtained by fitting the time complexity of the above processing functions. Finally, graph partitioning is adopted to divide the buffer task, which contributes to multiple parallel tasks matching with the computational resources.

  Results  The experimental results show that graph-based buffer analysis can achieve better load balance and overall efficiency, which is superior to the mainstream partitioning methods, regular-grid and quadtree.

  Conclusions  The proposed method can provide a reference for optimization of spatial analysis methods when processing large scale vector data.