不同尺度DEM的河流裂点提取及其效应分析

Identification of Fluvial Knickpoints and Analysis of Its Scale Effect Based on DEMs

以不同尺度DEM数据提取裂点及其效应存在较大差异。本文以1:1万DEM为基础数据,通过小波分析生成多尺度DEM数据。以庐山地区16条河流为例,实现了多尺度DEM数据的河流裂点提取,探讨了河流裂点的变化规律,并构建了裂点个数的尺度预测模型。实验结果表明:(1)采用河道纵剖面与点坡降相结合的方法可快速准确地判断裂点;(2)在庐山地区,1:1万DEM数据可准确判断高差不小于5 m的裂点,对于高差小于5 m的裂点由于DEM表达精度和数据误差,而无法准确判定;(3)DEM尺度对裂点提取影响显著,裂点个数随着DEM分辨率降低逐渐减少,符合幂函数递减规律;(4)通过与ASTER GDEM和SRTM DEM对比验证,本文所构建的裂点个数与DEM尺度的拟合模型具有一定的预测精度。

Knickpoints are fundamental for understanding local erosion basis and the evolution of fluvial land-forms. To extract the knickpoints, Digital Elevation Model （DEM） is widely adopted as the basic data in litera-tures. However, the accuracy of the extraction is greatly influenced by the DEM resolution. In this paper, to ex-plore the influence of DEM resolution on the extraction of knickpoints, we analyzed the gradient of the fluvial longitudinal profiles to extract the knickpoints in Mount Lu area based on DEM and Digital Line Graphic （DLG）. Firstly, the longitudinal profiles of the 16 streams with elevations are derived from 5 m DEM, from which the potential knickpoints are extracted from an empirical gradient domain. Secondly, to find a suitable gra-dient domain, a field investigation of four typical rivers, including Three-Step Spring and Crane Ravine, is car-ried out to collect the spatial positions of 30 knickpoints with GPS. Thirdly, multiple resolutions of DEM data are generated by wavelet transformation based on the 5 m DEM. The knickpoints in each scale are extracted in the same way from the corresponding DEM. Finally, the influence of the DEM resolution on the accuracy of knickpoints is discussed. The experiment results reveal that the suitable threshold of gradient domain for 5 m DEM is 0.6, by which the knickpoints’altitude differences are greater than 5 m. The number of knickpoints is subject to a power function of the DEM resolution. As the DEM resolution coarsens, the number of knickpoints decreases. To verify this power function, the ASTER GDEM and SRTM DEM data are used, which reveals the consistency between the extracted knickpoints from the DEMs and the predicted ones from the function.