A common method of channel network extraction from digital elevation model (DEM) data is based on specification of a threshold area A(th), that is, the minimum support area required to drain to a point for a channel to form. Current efforts to predict A(th) from DEM data are inconclusive, and usually an arbitrary constant A(th) value is chosen for channel network extraction. In this paper, we study the effects of threshold area selection on the morphometric properties (such as drainage density, length of drainage paths, and external and internal links) and scaling properties (such as Horton's laws and fractal dimension) of a channel network. We also study the related problem of DEM data resolution and its effect on estimation of scaling properties. The results indicate that morphometric properties vary considerably with A(th), and thus values reported without their associated A(th) are meaningless and should be used in hydrologic analysis with caution. Also, the ''completeness'' of a channel network (in terms of having the outlet stream flowing directly into a higher-order stream) is found to depend on A(th) in a random unpredictable way. Even if only the complete channel networks are used in the analysis, the statistical variability of scaling properties estimates due to A(th) selection is significant and can be of comparable size to the variability due to DEM resolution and variability between estimates of different river networks. Our analysis highlights the need to carefully study the problem of network source representation or channel initiation scale from DEMs which will point to an appropriate A(th) for channel network extraction and estimation of morphometric properties.
