On-the-fly analysis of multi-dimensional rasters in a GIS

Abstract

Geographic Information Systems and other mapping applications that specialize in image analysis routinely process high-dimensional gridded rasters as multivariate data cubes. Frameworks responsible for processing image data within these applications suffer from a combination of key shortcomings: inefficiencies stemming from intermediate results being stored on disk, the lack of versatility from disparate tools that don't work in unison, or the poor scalability with increasing volume and dimensionality of the data. We present raster functions as a powerful mechanism for processing and analyzing multi-dimensional rasters designed to overcome these crippling issues. A raster function accepts multivariate hypercubes and processing parameters as input and produces one output raster. Function chains and their parameterized form, function templates, represent a complex image processing operation constructed by composing simpler raster functions. We discuss extensibility of the framework via Python, portability of templates via XML, and dynamic filtering of data cubes using SQL. This paper highlights how ArcGIS employs raster functions in its mission to build actionable information from science and geographic data—by shrinking the lag between the acquisition of raw multi-dimensional raster data and the ultimate dissemination of derived image products. ArcGIS has a mature raster I/O pipeline based on GDAL, and it manages gridded multivariate multi-dimensional cubes in mosaic datasets stored within a geodatabase atop an RDBMS. Bundled with raster functions, we show those capabilities make possible up-to-date maps that are driven by distributed geoanalytics and powerful visualizations against large volumes of near real-time gridded data.