Title: GeoPAT - Pattern-based GIS Software for Understanding Content of Large Earth Science Datasets
Presenting Author: Tomasz Stepinski
Organization: University of Cincinnati
Co-Author(s): Pawel Netzel, Jaroslaw Jasiewicz

Abstract:
GeoPAT (Geospatial Pattern Analysis Toolbox) is software that performs regionalization of very large Earth Science datasets based on similarity of spatial or temporal patterns. For example, using high resolution, U.S.-wide land cover data as an input GeoPAT finds a set of characteristic landscapes (patterns of land cover) and segments the U.S. into units of uniform patterns of land cover. Similarly, using topographic data as an input, GeoPAT identifies exemplars of topographic landscapes (patterns of landforms) and delineates physiographic units. Applying GeoPAT to a very high resolution RGB image of an urban area makes possible identification of different urban structure types. Fitting GeoPAT with world-wide climate data results in a global climate classification. GeoPAT was originally developed as an extension to GRASS GIS. This had limited it usability, so we are currently working on the free standing version of the software (GeoPAT 2.0) that would run natively on Linux, Windows, and OS X. We are also working on full parallelization of GeoPAT code for Xeon Phi. In pattern-based GIS the pixel-based input is divided into a macro grid of motifels; motifel is a square-shaped multi-pixel region encapsulating a local pattern (a motif) of input variable. Subsequent analysis (segmentation, clustering etc.) is performed on motifels. GeoPAT includes all tools required to perform pattern-based GIS: tiling of input into a grid of motifels, calculating pattern representation within motifels, calculation similarity between motifels, segmentation of the grid of motifels into homogeneous segments, operation on segments. etc. In addition, GeoPAT supports unsupervised and supervised analysis including search based on the query-by-example. Using GeoPAT as calculation engine, Web-based applications to search for similar patterns of land cover, physiography, and climate were developed and are available at http://sil.uc.edu