The Impact of Contributor Confidence, Expertise and Distance on the Crowdsourced Land Cover Data Quality

Abstract

There is much interest in the opportunities for formal scientific investigations afforded by crowdsourcing and citizen sensing activities. However, one of the critical research issues relates to the 'quality' of the data collected in this way. This paper uses volunteer data on land cover collected under the Geo-Wiki system, where contributors label the land cover class at a series of locations, with expert labels at the same locations. It examines the statistical relationships between the accuracy of volunteer labels, their self assessed confidence in labelling, their 'experiential distance' to the location under consideration and the level of their domain expertise. The results show that distance has a minor effect on the reliability of land cover labelling, and that generally expertise has a greater effect, but not for all land cover classes.