Volume XLII-3
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-3, 87-92, 2018
https://doi.org/10.5194/isprs-archives-XLII-3-87-2018
© Author(s) 2018. This work is distributed under
the Creative Commons Attribution 4.0 License.
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-3, 87-92, 2018
https://doi.org/10.5194/isprs-archives-XLII-3-87-2018
© Author(s) 2018. This work is distributed under
the Creative Commons Attribution 4.0 License.

  30 Apr 2018

30 Apr 2018

A FREE AND OPEN SOURCE TOOL TO ASSESS THE ACCURACY OF LAND COVER MAPS: IMPLEMENTATION AND APPLICATION TO LOMBARDY REGION (ITALY)

G. Bratic, M. A. Brovelli, and M. E. Molinari G. Bratic et al.
  • Politecnico di Milano, Department of Civil and Environmental Engineering, Piazza Leonardo da Vinci 32, 20133 Milan, Italy

Keywords: Global Land Cover, Accuracy Assessment, Free and Open Source, Error Matrix, Accuracy Measures

Abstract. The availability of thematic maps has significantly increased over the last few years. Validation of these maps is a key factor in assessing their suitability for different applications. The evaluation of the accuracy of classified data is carried out through a comparison with a reference dataset and the generation of a confusion matrix from which many quality indexes can be derived. In this work, an ad hoc free and open source Python tool was implemented to automatically compute all the matrix confusion-derived accuracy indexes proposed by literature. The tool was integrated into GRASS GIS environment and successfully applied to evaluate the quality of three high-resolution global datasets (GlobeLand30, Global Urban Footprint, Global Human Settlement Layer Built-Up Grid) in the Lombardy Region area (Italy). In addition to the most commonly used accuracy measures, e.g. overall accuracy and Kappa, the tool allowed to compute and investigate less known indexes such as the Ground Truth and the Classification Success Index. The promising tool will be further extended with spatial autocorrelation analysis functions and made available to researcher and user community.