Exploring the potential role of volunteer citizen sensors in land cover map accuracy assessment

Exploring the potential role of volunteer citizen sensors in land cover map accuracy assessment
Giles M. Foody and Doreen S. Boyd

School of Geography, University of Nottingham, Nottingham, NG7 2RD, UK (giles.foody@nottingham.ac.uk, doreen.boyd@nottingham.ac.uk)

Abstract: Two sources of volunteered geographical information are used to provide reference data on forest to inform the evaluation of the accuracy of the Globcover land cover map for West Africa. Ground based photographs acquired through an internetbased collaborative project were interpreted by a further set of four volunteers to provide reference data on forest. Little agreement was found between a set of four volunteer interpreters but the set of labels they generated enabled a latent class model to be used to estimate forest cover. The latter is a key environmental variable as well as a basic indicator of accuracy on a non site-specific basis. The lack of gold standard reference makes detailed assessment difficult but the close correspondence between the model derived estimate of forest cover and that depicted in the map suggest the approach, based on volunteered data, may have value in accuracy assessment.

Keywords: remote sensing, citizen sensing, neogeography, crowd-sourcing, accuracy, latent class model, forest, Globcover.

AttachmentSize
FoodyAccuracy2012.pdf96.04 KB