沟谷特征点是反映沟谷地貌空间形态分布的重要点位,也是研究沟谷地貌演化过程与机理的关键要素。因此,对不同沟谷特征点的有效提取,是沟谷形态研究的重要基础。本文采用DEM及其在水文分析中的多种衍生数据,通过流向追踪、邻域特征判断等一系列方法,实现对径流节点、径流源点、汇流源点、潜在裂点与流域出口点的快速、准确提取。同时,对沟谷特征点进行有效分级,本文基于Strahler河流分级法建立了相应的分类标准,对沟谷特征点进行了自动分类。通过使用陕西省宜君典型样区5 m分辨率的DEM数据进行实验,发现新算法计算效率高,结果准确,对潜在裂点也进行了有效探测,验证了算法的有效性。最后,对特征点提取的数据影响进行了详细分析。

Gully feature points (GFPs) are crucial for studies on spatial pattern and evolution of gully landforms, and the extraction of the GFPs is the basis of related researches. Previously, scholars have made different methods for GFP detection and most are about runoff nodes. However, these methods are still not integral and have deficiencies in either accuracy or computational efficiency as analyzed in this paper. Thus, based on detailed analysis of DEMs and their derivatives, a series of new algorithms are proposed to improve the performance of GFP detection (including runoff nodes, river heads, confluence origins, potential knickpoints and outlets) by trace of flow direction and neighborhood feature judgment. The methods examine the essential features of GFPs and most thresholds used are invariable in different kinds of landforms. Thus, both the robustness and efficiency of the extraction are improved by the new algorithms. As orders of nodes and knickpoints are important in further analyses, criteria for automatic order classification are also established based on Strahler rules. By using 5m resolution DEMs in the test area in Yijun County, Shaanxi Province, experiments are made to test the methods' ability in both GPF extraction and classification. The results of runoff nodes, river heads, confluence origins and outlets are consistent with manual marks. The detected potential knickpoints also create favorable conditions for researches on gully evolution and guarantee the integrity of GFPs to the maximum extent. Preliminary analyses are made using the GFP results and reflect the variation characteristics of the sum of GFPs as rainfall changes. Finally, the impact of watershed completeness on the accuracy of GFP extraction is discussed in details. It shows that both the loss of stream pixels and confluence accumulations could cause the omission of the points. As the influences brought by data incompleteness can hardly be removed in the extraction process, it is crucial to guarantee that the boundaries of watersheds are within the data extent.