Accuracy analysis and validation of global 1 km land cover data sets

An approach for analyzing and validating thematic and positional accuracy of high resolution global land cover data set is described. Portions of the International Geosphere/Biosphere Programme´s (IGBP) 1 Kilometer Land Cover Data Set are used as a test bed to evaluate procedures for accuracy assessment and validation. The 1 km Land Cover Data Set contains 17 land cover classes ranging from natural vegetation to developed land and non-vegetated classes. This study focuses on developing statistical map accuracy procedures for data set validation through a continental scale cross validation exercise using the 1 km land cover data set. The objective of this exercise is to assess the potential of the IGBP fast track validation procedures. These include implementation of an integrated overall map accuracy analysis as well as a detailed category specific thematic accuracy assessment for both a North American and African study site. Specific goals of this study are to: (1) determine whether different classification techniques employed for Africa and North America significantly affect the classification of the 1 km Land Cover Data Set; (2) determine whether systematic differences are introduced into the datasets by the separate processing flows; (3) examine and attempt to explain any non-systematic differences which may be identified, and; (4) derive a statistical estimate of these differences between the two datasets for each of the 17 Land Cover classes-or as many of the classes as are adequately represented