Accuracy assessment of supervised classification of submersed macrophytes: the case of the Gavaraget region of Lake Sevan, Armenia

The necessity of assessing the accuracy of spatial data derived from remote sensing methods and used in geographic information system (GIS) analyses has been regarded as a critical component of many projects. In this article, supervised classified QuickBird satellite imageries of submersed macrophytes of the Gavaraget region of Lake Sevan (Armenia) during 2006–2008 are validated in a GIS environment. The results of this assessment are represented by error matrices presenting the overall accuracy, the user and producer accuracies in each category, as well as the kappa coefficients. The assessments were done at three levels: the vegetation types, the growth (vertical) type, and the species levels. At the vegetation level, an overall accuracy of more than 86% was achieved in 2006 and 2007, while that in 2008 was about 72%. Substantial agreements were achieved between the classified remote sensing data and the reference (groundtruth) data for both 2006 and 2007. Only a fair agreement was attained in 2008. The kappa coefficients ranged from 0.39 to 0.89. At the growth type level, only the 2006 classification attained a substantial agreement, with an overall accuracy of 80%. The 2007 and 2008 classifications had above average and average overall accuracies of 72 and 64%, respectively. The kappa coefficient had a lowest value of 0.35 and a highest value of 0.70 at the growth type level. At the species level, Chara spp. or Zannichellia palustris and Potamogeton pectinatus had high user accuracies of more than 90% and a high overall accuracy of 82% in 2006. High user accuracies (>80%) occurred for P. pectinatus (for 2007 and 2008) and No Vegetation (for 2007). The remaining species, however, had low user and producer accuracies of less than 64%. The kappa coefficients ranged between 0.36 and 0.73 at the species level.