空间分辨率与取样方式对DEM流域特征提取的影响

The Effect of Spatial Resolution and Sampling Method on the Watershed Features Derived from DEM

随着数字水文的兴起和分布式水文模型研究的发展,利用DEM提取水文特征,进而进行水文模拟的方法越来越广泛地为水文学者所采用.空间分辨率的改变与DEM重新取样方式对水文模拟都会产生重要影响.采取不同取样方法获得多种尺度的DEM,对不同分辨率下的流域特征值进行了统计分析与比较,引入熵的概念度量不同分辨率的DEM包含的信息量,以及不同取样方式对信息量的影响.并计算了以50mDEM所包含的信息量为基准,在不同的信息损失下所要求的最低分辨率.

With the development of digital hydrology and distributed hydrological modeling, more and more hydrologists turn to h-ydrological modeling based on the watershed features derived from digital elevation model (DEM). The DEM spatial resolution and sampling method have great impact on hydrology modeling. In this paper the impact of the DEM resolution on derived drainage features is examined for a study basin using DEM of increasing grid size. The result shows that watershed features derived from different resolutions of DEM data may be statistically different. The watershed features values are statistically analyzed, such as elevation, gradient, length of watershed network, topographic index, watershed area, stream density and mainstream length. Some mean, various, maximal and minimal values of the items are analyzed in detail. The lower the resolution is, the smoother the terrain is. Average gradient and length of watershed network and stream density minish with decreasing DEM resolution. Topographic index increases with DEM resolution. The concept of entropy has been considered a promising method in this study as it quantitatively measures the produced information of an object (watershed). With the DEM grid size increasing, entropy becomes smaller and smaller; entropy of average sampling is larger than that sampled at the center. DEMs with various resolutions are obtained from two methods average sampling and sampling at the center. The result also shows that mean gradient of average sampling is smaller than that of sampling at the center; but the length of watershed network and stream density are on the other way round. This means the conflux time of average sampling may be longer than that of sampling at the center. And mainstream length of average sampling is more close to the original 50 m DEM one s. Some watershed features have good relation with entropy. Also, suggestions are made for calculating the lowest resolution of various errors based on different resolutions. The result shows the DEM at 222 m resolution approaching the DEM at 50 m resolution within 5% error, and the DEM at (1 000) m resolution approaching the DEM at 50 m resolution within 30% error.