## TerraStream: From elevation data to watershed hierarchies

### Cached

### Download Links

- [biogeometry.duke.edu]
- [www.cs.duke.edu]
- [www.cs.duke.edu]
- [www.cs.au.dk]
- [www.daimi.au.dk]
- [www.cs.duke.edu]
- [www.cs.duke.edu]
- DBLP

### Other Repositories/Bibliography

Venue: | Proc. ACM Sympos. on Advances in Geographic Information Systems |

Citations: | 11 - 7 self |

### BibTeX

@INPROCEEDINGS{Danner_terrastream:from,

author = {Andrew Danner and Pankaj K. Agarwal and Ke Yi and Lars Arge},

title = {TerraStream: From elevation data to watershed hierarchies},

booktitle = {Proc. ACM Sympos. on Advances in Geographic Information Systems},

year = {},

pages = {212--219}

}

### OpenURL

### Abstract

We consider the problem of extracting a river network and a watershed hierarchy from a terrain given as a set of irregularly spaced points. We describe TerraStream, a “pipelined ” solution that consists of four main stages: construction of a digital elevation model (DEM), hydrological conditioning, extraction of river networks, and construction of a watershed hierarchy. Our approach has several advantages over existing methods. First, we design and implement the pipeline so each stage is scalable to massive data sets; a single non-scalable stage would create a bottleneck and limit overall scalability. Second, we develop the algorithms in a general framework so that they work for both TIN and grid DEMs. Terra-Stream is flexible and allows users to choose from various models and parameters, yet our pipeline is designed to reduce (or eliminate) the need for manual intervention between stages. We have implemented TerraStream and present experimental results on real elevation point sets that show that our approach handles massive multi-gigabyte terrain data sets. For example, we can process a data set containing over 300 million points—over 20GB of raw data—in under 26 hours, where most of the time (76%) is spent in the initial CPU-intensive DEM construction stage. 1

### Citations

539 |
The input/output complexity of sorting and related problems
- Aggarwal, Vitter
- 1988
(Show Context)
Citation Context ...cases the transfer of data between disk and main memory, not CPU computation time, often becomes the primary bottleneck. Therefore we are interested in designing efficient algorithms in the I/O-model =-=[4]-=-. In this model, the machine consists of a main memory of size M and an infinite-size disk. A block of B consecutive elements can be transferred between main memory and disk in one I/O operation (or s... |

320 | External Memory Algorithms and Data Structures: Dealing with Massive Data
- Vitter
- 1981
(Show Context)
Citation Context ...of the number of I/Os it performs. Many fundamental problems have been solved in the I/O model. For example, sorting N elements requires SORT(N) = Θ( N B logM/B N ) I/Os. Refer to surveys by B Vitter =-=[31]-=- and Arge [5] for other results. However, only recently have terrain problems been considered in the I/O-model. I/O-efficient algorithms have been developed for construction of either TIN or grid DEMs... |

235 | Topological persistence and simplification
- Edelsbrunner, Letscher, et al.
(Show Context)
Citation Context ...nly for grids and assumes that all sinks fit in memory. This assumption does not hold for large high-resolution terrains. We instead use a partial flooding algorithm, based on topological persistence =-=[16, 15]-=-, that detects and removes only insignificant sinks,s(a) (b) (c) Figure 3. (a) Original terrain. (b) Terrain flooded with τ =∞. (c) Terrain partially flooded with persistence threshold τ = 30. as indi... |

174 | Geometry and Topology for Mesh Generation - Edelsbrunner - 2001 |

173 | External-memory graph algorithms
- Chiang
- 1995
(Show Context)
Citation Context ...ing r(u) along all outgoing paths starting from u, we can assign each vertex in a subtree of u the proper raise elevation. This traversal can be performed in O(SORT(N)) I/Os using standard techniques =-=[11, 6]-=-. What remains is to compute the raise elevations for all non-sink vertices in the height graph G. To do so we first assign a sink label to each vertex in G. A vertex u is assigned sink label v if the... |

172 | Constrained Delaunay triangulations
- Chew
- 1987
(Show Context)
Citation Context ... the TIN to match breakline segments and preserve important topological features. In TERRASTREAM, we use a randomized I/O-efficient algorithm [2] for constructing a constrained Delaunay triangulation =-=[11]-=- of a set S of N points and a set L of K line segments, where all K line segments appear as edges of the final triangulation. The algorithm uses SORT(N) expected I/Os if the number of constraining seg... |

104 | Extracting Topographic Structure from Digital Elevation Data for Geographic Information System Analysis”, Photogrammetric Engineering and Remote
- Jenson, Domingue
- 1988
(Show Context)
Citation Context ...flow modeling stage by removing insignificant sinks, while preserving significant sinks. The most widely used hydrological conditioning algorithm removes all sinks using a so-called flooding approach =-=[20]-=-, which simulates uniformly pouring water on the terrain until a steadystate is reached. A weakness of this approach is that it removes even significant sinks. See Figures 3(a) and (b). Furthermore, t... |

93 |
The extraction of drainage networks from digital elevation data, Computer vision, graphics and image processing
- O’Callaghan, Mark
- 1984
(Show Context)
Citation Context ...her in Section 4.2. We have implemented extensions of SFD and MFD models that incorporate routing on flat areas. Flow accumulation. Given a flow graph F(G) with flow directions, the flow accumulation =-=[26]-=- phase intuitively computes the area of the terrain represented by vertices upslope of each node v. More precisely, each vertex v in the flow graph F(G) is assigned some initial flow. Each vertex then... |

79 | External memory data structures
- Arge
- 2002
(Show Context)
Citation Context ...of I/Os it performs. Many fundamental problems have been solved in the I/O model. For example, sorting N elements requires SORT(N) = Θ( N B logM/B N ) I/Os. Refer to surveys by B Vitter [31] and Arge =-=[5]-=- for other results. However, only recently have terrain problems been considered in the I/O-model. I/O-efficient algorithms have been developed for construction of either TIN or grid DEMs from a set o... |

55 |
The buffer tree: A technique for designing batched external data structures
- Arge
(Show Context)
Citation Context ...ing r(u) along all outgoing paths starting from u, we can assign each vertex in a subtree of u the proper raise elevation. This traversal can be performed in O(SORT(N)) I/Os using standard techniques =-=[11, 6]-=-. What remains is to compute the raise elevations for all non-sink vertices in the height graph G. To do so we first assign a sink label to each vertex in G. A vertex u is assigned sink label v if the... |

47 | Hierarchical Morse complexes for piecewise linear 2-manifolds
- Edelsbrunner, Harer, et al.
- 2001
(Show Context)
Citation Context ...nly for grids and assumes that all sinks fit in memory. This assumption does not hold for large high-resolution terrains. We instead use a partial flooding algorithm, based on topological persistence =-=[16, 15]-=-, that detects and removes only insignificant sinks,s(a) (b) (c) Figure 3. (a) Original terrain. (b) Terrain flooded with τ =∞. (c) Terrain partially flooded with persistence threshold τ = 30. as indi... |

41 |
A new method for the determination of flow directions and contributing areas in grid digital elevation models
- Tarboton
- 1997
(Show Context)
Citation Context ...f v is selected. • Multi-flow-directions (MFD) model: for each vertex v, all edges from v to neighbors of lower height are selected. Several other flow-direction models have also been proposed (e.g., =-=[28, 21]-=-), and most of them can be incorporated in our pipeline. We refer the reader to [8] for more information on SFD and MFD routing. If the height of every vertex in G is distinct, we can easily construct... |

34 |
Streaming computation of Delaunay triangulations
- Isenburg, Liu, et al.
(Show Context)
Citation Context ...lts. However, only recently have terrain problems been considered in the I/O-model. I/O-efficient algorithms have been developed for construction of either TIN or grid DEMs from a set of input points =-=[1, 2, 12, 18, 19]-=-, as well as for certain water flow problems, including river network extraction, on grid DEMs [8]. I/O-efficient flow algorithms for grid DEMs have been distributed in the TERRAFLOW software package ... |

31 | I/O-efficient algorithms for problems on grid-based terrains
- Arge, Toma, et al.
(Show Context)
Citation Context ...coming flow and distributes flow to each downslope neighbor u with an edge (v, u) in F(G) using a given function. Our algorithm is a slight generalization of a O(SORT(N)) I/O algorithm by Arge et al. =-=[10]-=- developed for grid DEMs and implemented in TERRAFLOW. Given the flow accumulations for all vertices, we can extract river networks [26] simply by extracting edges incident to vertices whose flow accu... |

28 |
Terrain Analysis: Principles and Applications
- Wilson, Gallant
- 2000
(Show Context)
Citation Context ...e common in many GIS applications. Terrain modeling and analysis has been studied extensively in many different communities, and algorithms have been developed for many fundamental problems. Refer to =-=[32]-=- and the references therein for a survey. Many GIS applications use a pipeline (or work-flow) approach for combining many smaller, simpler algorithms into a larger, more complex application. Often, th... |

27 |
The assignment of drainage directions over flat surfaces in raster digital elevation models
- Garbrecht, Martz
- 1997
(Show Context)
Citation Context ... Recently, a new more realistic flat area routing approach was proposed by Soille et al. [27]. Their approach, based on geodesic time and distance, improves an earlier approach by Garbrecht and Martz =-=[17]-=-. Given a flat area, define H to be the set of flat vertices having an upslope neighbor. The algorithm of Soille et al. [27] computes the minimum distance dv from each of the other flat vertices v to ... |

26 | Randomized external-memory algorithms for some geometric problems
- Crauser, Ferragina, et al.
- 2001
(Show Context)
Citation Context ...lts. However, only recently have terrain problems been considered in the I/O-model. I/O-efficient algorithms have been developed for construction of either TIN or grid DEMs from a set of input points =-=[1, 2, 12, 18, 19]-=-, as well as for certain water flow problems, including river network extraction, on grid DEMs [8]. I/O-efficient flow algorithms for grid DEMs have been distributed in the TERRAFLOW software package ... |

18 | I/O-efficient batched union-find and its applications to terrain analysis
- Agarwal, Arge, et al.
- 2006
(Show Context)
Citation Context ... in which u is the parent of v. The merge tree has the property that the heights of vertices on any root-to-leaf path increase, while the persistence values decrease along such a path. Agarwal et al. =-=[3]-=- developed an O(SORT(N))-I/O algorithm for computing the persistence of all sinks in a triangular planar height graph, and computing the merge tree. They also developed and implemented a simpler and p... |

16 |
Spatial interpolation
- Mitas, Mitasova
- 1999
(Show Context)
Citation Context ...on approach for constructing a grid DEM of a user-specified cell size from S is to use one of many interpolation or approximation methods to compute a height value for each grid point (refer to e.g., =-=[22]-=- and the references therein). For inputs with more than a few thousand points, applying an interpolation method directly is infeasible because of the computational complexity of solving large systems ... |

15 |
I/O-efficient construction of constrained Delaunay triangulations
- Agarwal, Arge, et al.
- 2005
(Show Context)
Citation Context ...lts. However, only recently have terrain problems been considered in the I/O-model. I/O-efficient algorithms have been developed for construction of either TIN or grid DEMs from a set of input points =-=[1, 2, 12, 18, 19]-=-, as well as for certain water flow problems, including river network extraction, on grid DEMs [8]. I/O-efficient flow algorithms for grid DEMs have been distributed in the TERRAFLOW software package ... |

14 | Flow computation on massive grid terrains
- Arge, Chase, et al.
(Show Context)
Citation Context ... been developed for construction of either TIN or grid DEMs from a set of input points [1, 2, 12, 18, 19], as well as for certain water flow problems, including river network extraction, on grid DEMs =-=[8]-=-. I/O-efficient flow algorithms for grid DEMs have been distributed in the TERRAFLOW software package [8]. Very recently, an I/O-efficient algorithm for extracting watershed hierarchies from a grid DE... |

12 |
Automated recognition of valley lines and drainage networks from grid digital elevation models: a review and a new method
- Tribe
- 1992
(Show Context)
Citation Context ...r that was along the shortest Euclidean path along grid edges from v to the closest spill point. However, these approaches are not hydrologically realistic and tend to create many parallel flow lines =-=[29]-=-. Recently, a new more realistic flat area routing approach was proposed by Soille et al. [27]. Their approach, based on geodesic time and distance, improves an earlier approach by Garbrecht and Martz... |

12 |
A topological system for delineation and codification of the Earth’s river basins
- Verdin, Verdin
- 1999
(Show Context)
Citation Context ...regions, or watersheds, where all water flows towards a single outlet. Such a decomposition is the basis of several GIS algorithms for hydrological and pollutant transport modeling. Verdin and Verdin =-=[30]-=- described a Pfafstetter labeling method, which hierarchically divides a terrain into arbitrarily small regions, each with a unique label, such that the Pfafstetter labels encode topological propertie... |

10 |
Simultaneous spline interpolation and topographic analysis for lidar elevation data: methods for open source GIS
- Mitasova, Mitas, et al.
- 2005
(Show Context)
Citation Context ...near equations. We chose for TerraStream a recently developed I/O-efficient algorithm [1] that uses a quad-tree segmentation in combination with a regularized spline with tension interpolation method =-=[23]-=- to conh + SORT(T)) I/Os, where h is the struct a grid DEM in O( N B log M B height of a quad tree on S and T is the number of cells in the desired grid DEM. Our implementation is modular and allows u... |

10 | Carving and adaptative drainage enforcement of grid digital elevation models
- Soille, Vogt, et al.
- 2003
(Show Context)
Citation Context ...t. However, these approaches are not hydrologically realistic and tend to create many parallel flow lines [29]. Recently, a new more realistic flat area routing approach was proposed by Soille et al. =-=[27]-=-. Their approach, based on geodesic time and distance, improves an earlier approach by Garbrecht and Martz [17]. Given a flat area, define H to be the set of flat vertices having an upslope neighbor. ... |

9 | From point cloud to grid DEM: A scalable approach
- Agarwal, Arge, et al.
- 2006
(Show Context)
Citation Context |

9 |
An aspect driven kinematic routing algorithm
- Lea
- 1992
(Show Context)
Citation Context ...f v is selected. • Multi-flow-directions (MFD) model: for each vertex v, all edges from v to neighbors of lower height are selected. Several other flow-direction models have also been proposed (e.g., =-=[28, 21]-=-), and most of them can be incorporated in our pipeline. We refer the reader to [8] for more information on SFD and MFD routing. If the height of every vertex in G is distinct, we can easily construct... |

7 |
Flood Mapping Program. http://www.ncfloodmaps.com
- Carolina
(Show Context)
Citation Context ...rth lidar points (over 20GB of raw data) from the Neuse river basin in North Carolina as our main test data. This data is publicly available for download from the North Carolina flood mapping project =-=[25]-=- and covers an area of roughly 6,200 square miles (16,000 km 2 ). The average point spacing in the set is approximately 20 feet (6m). Pipeline scalability. We present three experiments on the Neuse ri... |

7 |
Open source GIS: A GRASS GIS Approach (3rd edition
- Neteler, Mitasova
- 2008
(Show Context)
Citation Context ...r, and flexible. It scales to single multi-gigabyte sized data sets, works for both grid and TIN DEMs, and is faster than other scalable algorithms currently implemented, e.g., TERRAFLOW [8] in GRASS =-=[24]-=-. The highly modular and configurable pipeline is designed to reduce manual intervention, and to allow for easy addition of new modeling features. Our approach provides several parameters to control t... |

6 | Generating raster DEM from mass points via TIN streaming
- Isenburg, Liu, et al.
- 2006
(Show Context)
Citation Context |

5 | I/O-efficient hierarchical watershed decomposition of grid terrain models
- Arge, Danner, et al.
- 2006
(Show Context)
Citation Context ...upstream and downstream ordering. At the topmost level the terrain is divided into nine disjoint watersheds; each of these watersheds are recursively divided into nine smaller watersheds. Arge et al. =-=[9]-=- previously developed an algorithm using O(SORT(N)+T/B) I/Os for computing the Pfafstetter labels of a grid DEM of N cells, where T is the total size of the labels. The algorithm uses a data structure... |

4 |
I/O Efficient Algorithms and Applications in Geographic Information Systems
- Danner
- 2006
(Show Context)
Citation Context ...y to condition the terrain for the purpose of flow modeling. To efficiently condition a terrain using partial flooding, we utilize the following property of the merge tree whose proof can be found in =-=[13]-=-: Let u be a node in the merge tree that does not correspond to a significant sink, but whose parent does. Let v be any node in the sub-tree rooted at u. Then the raise elevation r(v) of v is r(v) = r... |

1 |
Open source GIS: A GRASS GIS Approach, volume 773
- Neteler, Mitasova
- 2008
(Show Context)
Citation Context ...r, and flexible. It scales to single multi-gigabyte sized data sets, works for both grid and TIN DEMs, and is faster than other scalable algorithms currently implemented, e.g., TERRAFLOW [8] in GRASS =-=[24]-=-. The highly modular and configurable pipeline is designed to reduce manual intervention, and to allow for easy addition of new modeling features. Our approach provides several parameters to control t... |