基于凝聚聚类算法的重力异常区域划分

Region segmentation of gravity anomaly based on agglomerative clustering algorithm

重力异常平面图的区域划分是地球物理场及地质构造解释的重要参考,但对于大范围数据一直缺少自动化辅助划分方法。基于凝聚层次聚类算法,开展重力异常网格数据的自动化区域划分研究,以空间邻近性为限制条件,分析不同连接方法对模拟重力异常数据的聚类效果,最终选取以Ward法为连接方式的凝聚聚类法,对菲律宾海盆区域的简单布格重力异常数据进行应用实验,并与已有板块划分理论进行对比分析。结果表明:凝聚聚类算法能够根据重力异常数据的空间位置及其数值的大小进行较为合理的自动划分;对菲律宾海盆重力异常的聚类分析结果与已有的板块区划较为接近,能够反映出许多局部细节上的信息,有利于解释工作的开展。因此,聚类分析是研究地球物理场特征的有效辅助工具。

Region segmentation on gravity anomaly plane is important reference for interpretation of geophysical field and geological structure, but for large-scale data there are few approaches that could realize automatic auxiliary division.Based on hierarchical clustering, study on automatic region segmentation on grid data of gravity anomaly is performed.We added the connectivity constraint of spatial adjacency and analyzed results of different linkage criteria on synthetic gravity anomaly.Finally, we applied agglomerative clustering with linkage criterion of Ward, and tested it on simple Bouguer gravity of Philippine Basin area.The results are compared with the existing slab division theory and show that the agglomerative clustering could divide the gravity anomaly according to their spatial locations and values;the clustering results on data of Philippine Basin are close to the existing theory, and could reflect more details, which is helpful to further interpretation.Thus, clustering algorithms may be effective auxiliary means for study on features of geophysical field.