Completing fragmentary river networks via induced terrain

Fragmentary river segments have to be reconnected before addressing various routing and tracking problems. Elevation determines drainage directions, so the partial heights available through LiDAR may provide useful hints on how the segments should be joined. However, it is not trivial how this information can be applied. This paper bridges this gap by proposing the induced structure approach, which first approximates a terrain compatible with those observations, and then derives a river network from that induced terrain. Since the network is derived from an induced terrain that honors the partial observations, we expect that the derived river network will enforce most restrictions imposed by the partial observations. This paper also provides specifics on the implementation. In the first step regarding terrain reconstruction, we find that the optimal scheme depends on the height sample distribution. If the samples are sparsely yet evenly distributed, natural neighbor interpolation with stream burning (NN-SB) is the most cost-effective. If the samples are offered only at the given river locations, the hydrology-aware version of Over determined Palladian Partial Differential Equation (HA-ODETLAP) should be used instead. In the second step concerning river derivation, we find it necessary to favor those given river locations. Otherwise they will be missed out. We set their respective initial water amounts to the critical accumulation level to ensure a river flows across them. In the subsequent branch thinning process, those locations are protected from being trimmed. We foresee applications of our solution framework in a few 2D and 3D network tracing problems with similar observation distribution, like dendrite network reconstruction.