Multiscale curvatures for identifying channel locations from DEMs

• We developed a method called Hyperscale Channel Extraction (HCE).
• HCE identifies channel locations in a digital elevation model.
• HCE explicitly uses information at multiple scales simultaneously.
• HCE can distinguish noise and likely channelized regions.
• HCE counteracts the effect of blurring due to isotropic wavelet filtering.

Curvature based methods are suitable for channel identification in digital elevation models. One obstacle in using these methods is the fact that channels generally occur at multiple scales in the landscape, from small creeks to large rivers. In this paper, we show how likely channel pixels can be identified simultaneously at a range of scales using multiscale curvature operators applied to digital elevation models. Our proposed Hyperscale Channel Extraction (HCE) method localizes channels at the smallest scale while simultaneously tracking the shape of the channel at a full interval of scales (the hyperscale). We test the method using two different types of curvature, and apply and validate it to a catchment representing terrain with a high slope sampled by airborne laser altimetry. The test results demonstrate that by explicitly employing the extra dimension of scale to localize channels, (a) we are able to robustly identify channel pixels, as possible channel locations are tracked through a full interval of scales, (b) no more a priori determination of the relevant scale is necessary, and (c) only one parameter remains to be set: a threshold on the curvature value that has a clear physical interpretation.