High-resolution automated detection of headwater streambeds for large watersheds

Headwater streams, which are small streams at the top of a watershed, account for the majority of the total length of streams, yet their exact locations are still not well known. For years, many algorithms were used to produce hydrographic networks that represent headwater streams with varying degrees of accuracy. Although digital elevation models derived from lidar have significantly improved headwater stream detection, the performance of the algorithms on landscapes with different geomorphologic characteristics remains unclear. Here, we address this issue by testing different combinations of algorithms using classification trees. Homogeneous hydrological processes were identified through Quaternary deposits. The results showed that in shallow soil that mainly consists of till deposits, the use of algorithms that simulate the surface runoff process provides the best explanation for the presence of a streambed. In contrast, streambeds in thick soil with high infiltration rates were primarily explained by a small-scale incision algorithm. Furthermore, the use of an iterative process that simulates water diffusion made it possible to detect streambeds more accurately than all other methods tested, regardless of the hydrological classification. The method developed in this paper shows the importance of considering hydrological processes when aiming to identify headwater streams.