Original paper
Point of view: error estimation in field assignment of land-cover types
Eriksen, Eva L.; Ullerud, Heidrun A.; Halvorsen, Rune; Aune, Sigrun; Bratli, Harald; Horvath, Peter; Volden, Inger K.; Wollan, Anders K.; Bryn, Anders
Phytocoenologia Band 49 Heft 2 (2019), p. 135 - 148
79 references
published: Mar 29, 2019
published online: Nov 30, 2018
manuscript accepted: Oct 1, 2018
final revised version received: Sep 15, 2018
manuscript revision requested: Aug 10, 2018
manuscript received: Jun 19, 2018
Open Access (paper may be downloaded free of charge)
Abstract
Questions: Substantial variation between observers has been found when comparing parallel land-cover maps, but how can we know which map is better? What magnitude of error and inter-observer variation is expected when assigning land-cover types and is this affected by the hierarchical level of the type system, observer characteristics, and ecosystem properties? Study area: Hvaler, south-east Norway. Methods: Eleven observers assigned mapping units to 120 stratified random points. At each observation point, the observers first assigned a mapping unit to the point independently. The group then decided on a ‘true’ reference mapping unit for that point. The reference was used to estimate total error. ‘Ecological distance’ to the reference was calculated to grade the errors. Results: Individual observers frequently assigned different mapping units to the same point. Deviating assignments were often ecologically close to the reference. Total error, as percentage of assignments that deviated from the reference, was 35.0% and 16.4% for low and high hierarchical levels of the land-cover-type system, respectively. The corresponding figures for inter-observer variation were 42.8% and 19.4%, respectively. Observer bias was found. Particularly high error rates were found for land-cover types characterised by human disturbance. Conclusions: Access to a ‘true’ mapping unit for each observation point enabled estimation of error in addition to the inter-observer variation typically estimated by the standard pairwise comparisons method for maps and observers. Three major sources of error in the assignment of land-cover types were observed: dependence on system complexity represented by the hierarchical level of the land-cover-type system, dependence on the experience and personal characteristics of the observers, and dependence on properties of the mapped ecosystem. The results support the necessity of focusing on quality in land-cover mapping, among commissioners, practitioners and other end users. Taxonomic reference: Lid & Lid (2005) for vascular plants. Syntaxonomic reference: Halvorsen et al. (2015) for land-cover types. Abbreviations: ED = Ecological distance; GLM = Generalised linear model; LCE = Local complex environmental variable; NiN = Nature in Norway; TPI = Topographic position index.
Keywords
Between-observer variation • classification • consistency • environmental gradient • error • land-cover mapping • observer effect • point sampling • reliability • repeatability • vegetation ecology • vegetation mapping