Channel networks exert a key control on drainage basins shape and dynamics, including the transfer of water and sediments throughout basins, and thus hydrosedimentary hazards. Landscape dissection by channels results from the competition between hillslope processes and channelized erosion processes such as overland flow or debris flows. In contrast to fluvial channelization, the transition from hillslopes to debris-flow channels remain understudied, and high resolution LiDAR DEMs open new perspectives for the extensive extraction of channel heads. Several channel extraction methods exist but none is yet robust on fast eroding landscapes where debris flows occur. Here we develop the CO2CHAIN method which identifies the hillslope to channel transition in drainage basins based on relative changes of local and upstream measures of flow convergence. We calibrate CO2CHAIN by fitting its results to channel head mapping made by geomorphologists on four contrasted basins in the United States and France. Compared to state-of-the-art channel extraction methods, and without any recalibration, it achieves similar and higher accuracy in moderate and high erosion-rate basins, respectively. This allows to revisit studies on channelization that have not yet included debris-flow dominated basins, and better understand the characteristics and boundaries of the debris-flow domain.