Changing the TIGER's stripes: detecting road network change under positional uncertainty

Changing the TIGER's stripes: detecting road network change under positional uncertainty
Ashton Shortridge1 and Miaoying Shi2

1.Department of Geography, Michigan State University, USA 48824 (ashton@msu.edu)
2.Department of Forestry, Michigan State University, USA 48824 (miaoyingshi23@gmail.com)

Abstract: Many sets of linear features (e.g., streets in a city) can change over time, and the identification of these changes is an important geoprocessing challenge. This challenge is exacerbated by the often low accuracy of historic network datasets: positional discrepancies between features at different times may reflect actual change, or they may simply be the product of error. While there is a substantial body of work on modeling positional uncertainty in linear features, these contributions do not appear to have been widely integrated with feature change detection algorithms. In the present work, we address this research gap by developing several uncertainty modeling approaches for use in detecting significant network change. The following paragraphs review relevant approaches to characterizing positional uncertainty and then present a typology of network data mismatch. We then describe a geostatistical approach to model this uncertainty, and contrast it with conventional epsilon band technique to identify significant change. The approach is demonstrated in a case study with US Census TIGER line data

Keywords: positional error, vector uncertainty, kriging, epsilon bands

AttachmentSize
ShortridgeAccuracy2012.pdf457.45 KB