Variation in land cover classification due to individual interpreter assessment: a case study of farmland mapping in Norway |
| |
Authors: | Kjersti Straume |
| |
Institution: | 1. Department of Geography, University of Bergen, PB 7800, N-5020 Bergen, Norwaykjersti.straume@geog.uib.no |
| |
Abstract: | Visual interpretations of remote sensing images are required when producing land cover/use maps for management purposes. Human assessment is viewed as an advantage during the interpretation process. However, large mapping projects require several interpreters, and the possibility of inter-operator variability during classification work may affect map usage. In this study, five skilled interpreters from the Norwegian Forest and Landscape Institute independently updated the national Norwegian land resource dataset (AR5) in the same two study areas using the same orthophotos. Cross-tabulations were obtained, and class agreement, overall agreement, kappa agreement and fuzzy Kappa (Kfuzzy) were calculated. Most disagreements occurred in fragmented and complex landscapes, particularly with closely related classes. What can be regarded as an acceptable level of agreement depends on the intended map use and should be based on sensitivity tests exploring the effects of interpreter variation on specific applications of the map product. Finally, the threshold is a political decision based on thorough consideration of the economic and non-economic consequences of the variation tolerated in the maps produced. |
| |
Keywords: | assessment of agreement cross-tabulations GIS land cover classification mapping |
|
|