Proposed Research Priority

Conflation: Combining GIS Sources

Prepared by Michael Goodchild, Geography and NCGIA

other data acquisition and measurement topics (10.10), accuracy (10.1), pattern recognition (11.7), stochastic processes (4.7)

Description of topic

There are abundant statistical techniques for combining information from different sources and for determining the variation between them. The mean is often used to summarize a sample of observations, and the standard deviation to summarize variability. But no comparable, standard techniques exist for combining geographic information. The general term "conflation" is used in the GIS industry and elsewhere for the many versions of this problem, which range from combining digitized topographic maps and GPS as sources of street centerline locations, to edgematching of misfit data across boundaries, to combining information from different sensors in remote sensing.


Conflation research is under way in numerous agencies, institutions, and in the vendor community. But at this time it lacks a coherent framework, an underlying theory, and the benefits of a coordinated approach--it is not recognized as a generic problem in GIS. Research is essential if the concepts of partnerships embedded in NSDI are to be technically feasible, because they rely on the ability to integrate data from various sources across scales and between adjacent jurisdictions. Conflation of vector data with images will be increasingly important as more digital geographic data becomes available over the Web and through digital spatial data libraries. Conflation of images from different sensors allows value to be added by combining images of high spectral resolution with images of high spatial resolution.

Examples of research topics

Many fundamental questions need to be addressed through research. How many distinct versions of the conflation problem are there? Do solutions already exist within the relevant literatures of statistics and geometry? What progress has already been made in related fields such as image processing and CAD? How good are the conflation packages already available from software vendors? What data models would allow features to retain the legacy of their precursor positions? What is the role of semantics in conflation, versus geometry and topology? Is it possible to weight inputs, when information on relative reliability is available?


A two-year, dedicated effort could produce a national consensus on conflation in GIS that would be of great value to vendors, academics, and agencies.

If you have any suggestions for revision of the above, please send your comments to either Michael Goodchild or Karen Kemp.

 Go back to UCSB-UCGIS homepage

Posted February 28, 1996.