Swarthmore College Department of Computer Science

Talk by Andy Danner, Swarthmore College

TerraStream: From Elevation Data to Watershed Hierarchies
Monday, February 4, 2008
3:45 pm in Science Center 240
Reception and refreshments at 3:30 pm

Abstract

Modern remote sensing and mapping technologies generate Geographic Information Systems (GIS) that often exceed several Gigabytes or Terrabytes in size. Processing such huge data sets poses a number of computational challenges. Portions of the data must reside on large but slow hard disks, while computation can only occur in the smaller but faster internal memory of modern computers. In these cases the transfer of data between disk and main memory becomes the primary bottleneck rather than internal CPU computation.

This talk will describe the I/O model of computation in which we can develop scalable algorithms for processing large data sets. I will also present TerraStream--an implementation of several I/O-efficient algorithms for processing large point clouds of elevation data, creating digital surface models, extracting river networks, and constructing watershed hierarchies. TerraStream performance scales efficiently to input data sets containing over 300 million points and over 20GB in size.