Trust as a proxy indicator for intrinsic quality of Volunteered Geographic Information in biodiversity monitoring programs. Issue 4 (4th July 2018)

• Record Type: Journal Article
• Authors: Vahidi, Hossein; Klinkenberg, Brian; Yan, Wanglin
• Is Part Of: GIScience & remote sensing. Volume 55:Issue 4(2018)
• Page Start: 502
• Page End: 538
• Publication Date: 2018-07-04
• Subjects: Volunteered Geographic Information -- citizen science -- crowdsourcing -- intrinsic data quality indicator -- trust -- biodiversity monitoring; Geodesy -- Periodicals; Cartography -- Periodicals; Aerial photogrammetry -- Periodicals; Remote sensing -- Periodicals; 526.05
• DOI: 10.1080/15481603.2017.1413794
• Languages: English
• ISSNs: 1548-1603

Abstract

Abstract : In this article, we present a fuzzy model for intrinsic quality assessment of Volunteered Geographic Information (VGI) on species occurrences obtained by Citizen Science (CS) biodiversity monitoring programs. The proposed VGI quality assurance approach evaluates the thematic and positional quality of the crowdsourced biodiversity observation in terms of the trustworthiness of the observation by combining three indicators of consistency with habitat, consistency with surroundings, and reputation of contributor, that characterize the geographical and social aspects of trust in VGI. To evaluate the performance and usability of the proposed approach for evaluating the trustworthiness of crowdsourced observations and detecting thematic and positional errors in crowdsourced observations, the developed approach was applied to the crowdsourced observations on Acer macrophyllum collected through the CS biodiversity monitoring projects of E-Flora BC and iNaturalist. The result of a conformity test at the optimal acceptance threshold (sensitivity = 0.99, specificity = 0.8, and Cohen's kappa = 0.79), the achieved area under the curve (AUC) value (AUC = 0.98), and the results of the complementary investigation on the predictions of the proposed model indicated that the proposed fuzzy trust model exhibited promising predictive performance and was able to flag the majority of attribute and positional errors in the crowdsourced biodiversity observations.