Fully Automated Unstructured Mesh Generation Based on a Hierarchized Ridgeline Network

Abstract:

High resolution digital elevation models (DEMs) allow hydrologists to describe accurately the land surface topography. Hydrological models based on structured grids can hardly incorporate these descriptions due to prohibitive computational costs involved by the finest available resolution and the loss of details in critical areas involved by coarsening. In this study, we provide a new fully automated method for ridgeline delineation and hierarchy definition that allow us to generate unstructured meshes where the detail is retained only when it is important. Two networks are extracted: (1) the drainage network (DN) and (2) the ridgeline network (RN). The DN (cyan lines in Figure a) is extracted from high resolution DEMs by using the D8-LTD method without filling natural and artificial depressions. Cell vertices and mid edge cell points (red points in Figure b) are defined to be ridge points (RPs). These RPs are used to define the RN (green lines in Figures c and d). RP elevations are defined as the maxima among the neighboring cell center elevations. The DN allows us to define whether a RP belongs to a ridgeline within the same basin or to a ridgeline that separates two different (endorheic) basins. For each ridgeline that separates two (endorheic) basins the saddle displaying lower elevation is detected (dark blue points in Figure b). By considering saddles in the order of ascending elevation, the path from which flows can spill out from endorheic basins (dark blue lines in Figure b) can be identified. To obtain the RN, RPs are considered in the order of ascending elevation and joined sequentially to form the ridgeline. When two ridgelines meet, the one started from the lowest elevation continues and the other is considered to be a secondary ridgeline. The procedure stops when the highest RP is reached. For any grid resolution and terrain complexity, properly ordered DN and RN are extracted to obtain a couple of complementary networks that represent all the essential morphological features. These two networks can eventually be pruned and used to generate unstructured meshes (black lines in Figure d). These meshes can be used to develop reliable and computational efficient distributed catchment models or 2D floodplain models (see the AGU FM abstract ID# 669158).