Marcus V. A. Andrade

Parallel ODETLAP for terrain compression and reconstruction

We introduce a parallel approximation of an Over-determined Laplacian Partial Differential Equation solver (ODETLAP) applied to the compression and restoration of terrain data used for Geographical Information Systems (GIS). ODETLAP can be used to reconstruct a compressed elevation map, or to generate a dense regular grid from airborne Light Detection and Ranging (LIDAR) point cloud data. With previous methods, the time to execute ODETLAP does not scale well with the size of the input elevation map, resulting in running times that are prohibitively long for large data sets. Our algorithm divides the data set into patches, runs ODETLAP on each patch, and then merges the patches together. This method gives two distinct speed improvements. First, we provide scalability by reducing the complexity such that the execution time grows almost linearly with the size of the input, even when run on a single processor. Second, we are able to calculate ODETLAP on the patches concurrently in a parallel or distributed environment. Our new patch-based implementation takes 2 seconds to run ODETLAP on an 800 x 800 elevation map using 128 processors, while the original version of ODETLAP takes nearly 10 minutes on a single processor (271 times longer). We demonstrate the effectiveness of the new algorithm by running it on data sets as large as 16000 x 16000 on a cluster of computers. We also discuss our preliminary results from running on an IBM Blue Gene/L system with 32,768 processors.

Surface compression using over-determined Laplacian approximation

We describe a surface compression technique to lossily compress elevation datasets. Our approach first approximates the uncompressed terrain using an over-determined system of linear equations based on the Laplacian partial differential equation. Then the approximation is refined with respect to the uncompressed terrain using an error metric. These two steps work alternately until we find an approximation that is good enough. We then further compress the result to achieve a better overall compression ratio. We present experiments and measurements using different metrics and our method gives convincing results.

Efficient viewshed computation on terrain in external memory

The recent availability of detailed geographic data permits terrain applications to process large areas at high resolution. However the required massive data processing presents significant challenges, demanding algorithms optimized for both data movement and computation. One such application is viewshed computation, that is, to determine all the points visible from a given point p. In this paper, we present an efficient algorithm to compute viewsheds on terrain stored in external memory. In the usual case where the observer’s radius of interest is smaller than the terrain size, the algorithm complexity is θ(scan(n2)) where n2 is the number of points in an n × n DEM and scan(n2) is the minimum number of I/O operations required to read n2 contiguous items from external memory. This is much faster than existing published algorithms.

Smugglers and border guards: the GeoStar project at RPI

We present the GeoStar project at RPI, which researches various terrain (i.e., elevation) representations and operations thereon. This work is motivated by the large amounts of hi-res data now available. The purpose of each representation is to lossily compress terrain while maintaining important properties. Our ODETLAP representation generalizes a Laplacian partial differential equation by using two inconsistent equations for each known point in the grid, as well as one equation for each unknown point. The surface is reconstructed from a carefully-chosen small set of known points. Our second representation segments the terrain into a set of regions, each of which is simply described. Our third representation has the most long term potential: scooping, which forms the terrain by emulating surface water erosion. Siting hundreds of observers, such as border guards, so that their viewsheds jointly cover the maximum terrain is our first operation. This process allows both observer and target to be above the local terrain, and the observer to have a finite radius of interest. Planning a path so that a smuggler may get from point A to point B while maximally avoiding the border guards is our second operation. The path metric includes path length, distance traveled uphill, and amount of time visible to a guard. The quality of our representations is determined, not only by their RMS elevation error, but by how accurately they support these operations.

A New Method for Computing the Drainage Network Based on Raising the Level of an Ocean Surrounding the Terrain

We present a new and faster internal memory method to compute the drainage network, that is, the flow direction and accumulation on terrains represented by raster elevation matrix. The main idea is to surround the terrain by water (as an island) and then to raise the outside water level step by step, with depressions filled when the water reaches their boundary. This process avoids the very time-consuming depression filling step used by most of the methods to compute flow routing, that is, the flow direction and accumulated flow. The execution time of our method is very fast, and linear in the terrain size. Tests have shown that our method can process large terrains more than 100 times faster than other recent methods.

More efficient terrain viewshed computation on massive datasets using external memory

We present a better algorithm and implementation for external memory viewshed computation. It is about four times faster than the most recent and most efficient published methods. Ours is also much simpler. Since processing large datasets can take hours, this improvement is significant. To reduce the total number of I/O operations, our method is based on subdividing the terrain into blocks which are stored in a special data structure managed as a cache memory. The viewshed is that region of the terrain that is visible by a fixed observer, who may be on or above the terrain. Its applications range from visual nuisance abatement to radio transmitter siting and surveillance.

EXTENDING THE UML-GEOFRAME DATA MODEL FOR CONCEPTUAL MODELING OF NETWORK APPLICATIONS

This paper presents an extension of the UML-GeoFrame data model that includes a set of new constructors to allow the definition of conceptual schemas for spatial database applications whose elements relationship forms a network.. Also, it is discussed how the GeoFrame conceptual framework is changed with the inclusion of new metaclasses and the corresponding stereotypes related to network elements. The extension proposed in this paper is evaluated using a class diagram for a water distribution company.

Computing the drainage network on huge grid terrains

We present a very efficient algorithm, named EMFlow, and its implementation to compute the drainage network, that is, the flow direction and flow accumulation on huge terrains stored in external memory. It is about 20 times faster than the two most recent and most efficient published methods: TerraFlow and r.watershed.seg. Since processing large datasets can take hours, this improvement is very significant. The EMFlow is based on our previous method RWFlood which uses a flooding process to compute the drainage network. And, to reduce the total number of I/O operations, EMFlow is based on grouping the terrain cells into blocks which are stored in a special data structure managed as a cache memory. Also, a new strategy is adopted to subdivide the terrains in islands which are processed separately. Because of the recent increase in the volume of high resolution terrestrial data, the internal memory algorithms do not run well on most computers and, thus, optimizing the massive data processing algorithm simultaneously for data movement and computation has been a challenge for GIS.

Fast exact parallel map overlay using a two-level uniform grid

We present EPUG-Overlay (Exact Parallel Uniform Grid Overlay), an algorithm to overlay two maps that is fast and parallel, has no roundoff errors, and is freely available. EPUG-Overlay combines several novel aspects. It represents coordinates with rational numbers, thereby ensuring exact computations with no roundoff errors and the ensuing sliver problems and topological impossibilities. For efficiency, EPUG-Overlay performs the map overlay in parallel, thereby utilizing the ubiquitous multicore architecture. Our application goes beyond merely using existing packages, which are inefficient when used in parallel on large problems. Indeed, overlaying two maps with 53,000,000 edges and 730,000 faces took only 322 elapsed seconds (plus 116 seconds for I/O) on a dual 8-core 3.1 GHz Intel Xeon E5-2687 workstation. In contrast, GRASS, executing sequentially and generating roundoff errors, takes 5300 seconds. The overlay operation combines two input maps (planar graphs) containing faces (polygons) separated by polyline edges (chains), into a new map, each of whose faces is the intersection of one face from each input map. Floating point roundoff errors can cause an edge intersection to be missed or the computed intersection point be in a wrong face, leading to a topological inconsistency. Thus, a program might fail to compute a valid output map at all, using any amount of time. This gets worse when the inputs are bigger or have slivers. Heuristics can ameliorate this problem, but only to an extent. By representing each coordinate as a vulgar fraction, with multiprecision numerator and denominator, the computation is exact. EPUG-Overlay also executes various useful subproblems very quickly, such as locating a set of points in a planar graph and finding all the intersections among a large set of small edges. EPUG-Overlay is built on our earlier sequential floating-point algorithm that found the areas of the overlay polygons, without finding the polygons themselves.

An efficient GPU multiple-observer siting method based on sparse-matrix multiplication

This paper proposes an efficient parallel heuristic for siting observers on raster terrains. More specifically, the goal is to choose the smallest set of points on a terrain such that observers located in these points are able to visualize at least a given percentage of the terrain. This problem is NP-Hard and has several applications such as determining the best places to position (site) communication or monitoring towers on a terrain. Since siting observers is a massive operation, its solution requires a huge amount of processing time even to obtain an approximate solution using a heuristic. This is still more evident when processing high resolution terrains that have become available due to modern data acquiring technologies such as LIDAR and IFSAR. Our new implementation uses dynamic programming and CUDA to accelerate the swap local search heuristic, which was proposed in previous works. Also, to efficiently use the parallel computing resources of GPUs, we adapted some techniques previously developed for sparse-dense matrix multiplication. We compared this new method with previous parallel implementations and the new method is much more efficient than the previous ones. It can process much larger terrains (the older methods are restrictive about terrain size) and it is faster.