关键词: 地貌面, 沉积特征, 粒度分析, 自组织人工神经网络, 系统聚类分析

Abstract: When using river geomorphology to study tectonic deformation, it is often difficult to distinguish the same level geomorphology in areas with severe weathering. In this paper, we take the geomorphologic surfaces of the Qingyijiang river basin as an example and try to distinguish the geomorphic surfaces by the sediment features that make up them. In order to distinguish different geomorphic surfaces, the traditional particle-size analysis method, SOFM network method and system clustering analysis method are taken to classify 29 samples from different geomorphic surfaces. The classification results of the three methods are different to a certain extent. We analyzed and compared the classification results of the three methods in detail. The results show that the traditional particle size analysis method, SOFM network method and cluster analysis method all can distinguish the geomorphic surface of different genesis, besides, they also can distinguish low-level terraces(T₁, T₂)and high-level terraces(T₃, T₄)for different grades of river terraces. Furthermore, the results also show that SOFM network method and cluster analysis method can make a certain distinction for the low-level terraces(T₁, T₂), while the traditional particle size analysis method is difficult to distinguish them.
In addition, we analyzed and compared the three methods from the classification results, the results presentation, the operation process, and the error transmission. The results suggest that the advantages and disadvantages of the three methods are obvious. From the perspective of the classification results, the three methods all can distinguish the river terraces and alluvial fans and can make certain discrimination for different levels of river terraces. From the presentation of the results, the result of SOFM network is simple and clear. From the operation process, the traditional particle-size analysis method is relatively cumbersome, and the SOFM network method and the cluster analysis method are relatively simple to operate. From the perspective of error transmission, the traditional particle-size analysis method calculates the partial particle size feature value of the sample, which has a certain loss for the particle size distribution information of the whole sample. The error of the clustering analysis method has cumulative features and the influence exists consistently. The classification results of the SOFM network are independent of each other, which effectively avoids the problem of such error transmission of clustering analysis method.
Overall, the classification results of the SOFM network method are simple and clear, the operation is simple, and the error is small. It has stronger adaptability to identifying different levels of different geomorphic surfaces. The results of this study will provide a simple and effective means for distinguishing different levels of geomorphic surfaces.

Key words: geomorphic surfaces, sedimentary characteristics, particle size analysis, Self-Organizing Feature Map(SOFM), artificial neural network, system clustering analysis