Generalization Quality Metrics in the Age of Big Geospatial Data

Keywords: Big data, generalization, data quality assessment, cartography
Abstract Type: Paper Abstract

Abstract

Massive geospatial datasets collected from various sources at different frequencies and resolutions are readily available for download. Because of challenges associated with their use, vector and raster data must often be generalized for analysis and display at reduced scales. This paper describes metrics for assessing the quality of geospatial data after generalization operations. These metrics are needed for at least two reasons. First, users need to understand the effects of common generalization operations on geospatial data, analyses, and decision-making to ensure that processing has not obscured properties that are relevant to the purpose for downloading the data. Second, federal agencies producing geospatial datasets must provide easily accessible metadata describing essential information about a dataset, including associated quality information that can assist in assessing the data fitness for a particular use. Distribution of generalized datasets is limited by a lack of standardized metrics and tools to assess and report data quality. The proposed tools evaluate generalized data quality according to three broad categories. Positional accuracy metrics quantify maximum and average displacement between original and generalized feature positions. Cartographic visibility metrics quantify density and legibility of generalized features. Retention of characteristic properties, such as sinuosity, are assessed to preserve analytic fidelity. Automated computations of proposed metrics are implemented using open-source Python libraries, with the goal of creating an easy-to-use generalization assessment toolbox. Utilization of this toolbox and incorporation of metrics into metadata standards should help educate data users and control the quality of generalized data for analysis and cartographic purposes.