|
Land cover data derived from satellite images must be accompanied by information on their quality in order to be properly and fully used. The main purpose of this paper is to present the results of the evaluation of the Urban Atlas (UA) and Street Tree Layer (STL) datasets for Bulgaria for the reference 2012 year. This verification task is part of a project, managed by the European Environmental Agency (EEA) under the Copernicus program. A quantitative assessment approach is applied, based on stratified random sampling at polygon level. The working steps completed are, as follows: preparation of input and reference data, creation of a sample set of land cover/land use (LC/LU) polygons, visual interpretation of selected samples and evaluation of results. The LACO-Wiki web-based tool is used for sampling-related activities. Very highresolution (VHR) satellite imagery and digital aerial photos form major part of the reference data. This way, a scientifically based estimate of the thematic accuracy of UA and STL dataset are obtained and of some geometric characteristics, as well. The following features are estimated: user and producer's accuracy by LC/LU class, overall user's accuracy, uncertainty values, and correctness of delineation by LC/LU class. Comments by LC/LU class are provided. More than 89% of UA polygons have correctly delineated area. The detail of delineation accuracy is over 98%, while the positional accuracy is more than 97%. The overall weighted thematic accuracy of UA is 83.9%, which is higher than the target accuracy of 80%. STL product shows overall accuracy of 94.1% - higher than the required 85%. Relevant and diverse reference data sources together with appropriate stratification and sampling design tailored to the purpose and resources of the project helped to produce realistic accuracy results.
|
