Abstract. Headwaters make up the majority of any given stream network, yet, they are poorly mapped. A solution to this is to model the stream networks from a high resolution digital elevation model. Selecting the correct stream initiation threshold is key, but how do you do that on a national scale across physiographic regions? Here the Swedish landscape is used as a test bench to investigate how the mapping of small stream channels (< 6 m width) can be improved. The best modelled stream channel network was generated by pre-processing the DEM, calculating the accumulated flow, extracting a stream network using a stream initiation threshold of 2 ha. The Matthews Correlation Coefficient (MCC) for the 2 ha stream channel network was 0.463 while the best available maps of today, the Swedish property map (1 : 12 500) had an MCC of 0.387. A residual analysis of the 2 ha network show that there is additional improvements to be made by adapting the model to local conditions, as 15 % of the over and underestimation could be explained by the variability in runoff, quaternary deposits, local topography and location. The most accurate stream channel network had a length 4.5 times longer the currently mapped stream network, demonstrating how important accurate stream networks is for upscaling aquatic and climate research.