The flow directions of the extracted drainage network are more random based on random flow model. Binary linear regression method is applied to calculate the residual value at every point, and then the normalized residuals are used to prune reasonably as the variables. The terrains are distinguished automatically according to the normalized residuals, and the local catchment area threshold is determined based on each terrain feature. When there are two terrains in the experimental area, source density is large on mountainous terrain, and it is small on flat terrain. Eventually, the extracted result is consistent with the actual drainage network.
