Elsevier

Landscape and Urban Planning

Volume 78, Issue 4, 28 November 2006, Pages 289-299
Landscape and Urban Planning

Accuracy and congruency of three different digital land-use maps

https://doi.org/10.1016/j.landurbplan.2005.09.004Get rights and content

Abstract

For numerous model applications in the earth and environmental sciences, digital land-use data are indispensable as a source of information on the geographical distribution of the land-use/cover. Therefore, the land-use data sets ATKIS, CORINE Land Cover, and Landsat TM classifications are widely used in Germany. However, the users of these data mostly do not have information on their quality. In this study, the accuracy of the three above-mentioned digital land-use maps was evaluated a posteriori based on the results of a field inventory in two test areas with a total area of 17 km2. The results show that the overall accuracy of the land-use maps varies from 0.692 to 0.876. For the individual land-use classes, the user accuracy ranged from less than 0.001 to 0.991. In this paper, the positional congruency of the land-use class polygons among the three maps was also evaluated for a larger study region of 670 km2 in the state of Hessen (Germany). This region is a small structured landscape with a relatively high portion of fallow land. In the analysis, the following six land-use classes were considered: urban and traffic areas, forest, water, arable land, pastures and meadows, and fallow land (including other uses). The results showed that the congruency of the land-use classes forest and urban and traffic areas is higher than the congruency of the land-use classes of the open land (arable land, pastures and meadows, fallow).

Introduction

Process models describing biological, physical and chemical cycles, energy turnover, climate change, habitat suitability, or ecological hazards on different scales are among the most important scientific tools currently available to describe and analyze the status, pressures, and changes in the Earth's environmental systems. Information on the actual land-use/cover is necessary for the accurate description of many physical processes taking place on the earth surface. Thus, land-use/cover maps are used in numerous model applications to describe the spatial allocation and pattern of land-cover and to estimate aerial extent and location of various cover classes (Stehman and Czaplewski, 1998).

Meso- and macro-scale modeling tasks require us to revert back to digital data for the land-use distribution. For this purpose, digital land-use maps and remote sensing data are commonly used as a source of geographically referenced information on land-use and land-cover, respectively. The usefulness of digital land-use data sets for the user depends, among other factors, on the data accuracy, i.e., how precise these data describe the reality. With the increasing use of digital geographic information, the assessment of the accuracy and errors in digital land-use/land-cover maps and remote sensing products has received considerable attention in the literature (Edwards et al., 1998). For example, Bolstad and Smith (1992) discuss both positional and attribute accuracy, while Lunetta et al. (1991) discuss how errors propagate through the map-making process.

Producing data is one thing, assessing its quality another but essential for the data user. Unfortunately it has not yet become a standard to state the positional and thematic accuracies of a data set though it should be part of the metadata of every data set, especially in such large national data collection efforts as the CORINE Land Cover map and FAO AFRICOVER Project (FAO, 2003). Some projects are being formulated and executed successfully in which the overall thematic data accuracy is defined a priori (e.g., in World Bank projects). Also, an a posteriori approach to thematic data accuracy is possible; the 1-km Global Land-Cover data set of the International Geosphere-Biosphere Programme, Data and Information System (IGBP-DIS) is a good example at global and continental level that illustrates also the variety of accuracy levels within the data set (Loveland and Belward, 1997).

There are two primary motivations for assessing the accuracy of a map. The first motivation is to attempt to detect and understand the errors in the map. This is of interest to the producers of thematic maps, who are interested in this kind of information to improve the map-making process. Information about the errors in the map, in turn, may help users of the maps to interpret and use the maps more effectively. However, assessing the accuracy of a land-use map is a time-consuming and expensive process for larger areas. For land-use/land-cover maps generated from aerial photos or satellite images, the evaluation of database accuracy and the circulation of the results usually is an integral part of the production process (e.g., Chust et al., 2004, Congalton and Green, 1999, Goodchild and Gopal, 1989, Li et al., 2004). As a common criterion, the overall accuracy is used as an indication of the general reliability of a map, while the error matrix provides information on the reliability of the individual classes. In contrast to remote sensing based products, information on the quality and accuracy of commonly used digital topographic maps is sparse (Bender et al., 2005, Sbresny, 1997).

From the user perspective, the digital land-use map is not the final product, but input data are required for a model. The decision to use a specific land-use map as input is based on a range of factors like thematic content, scale, spatial resolution, classification system, aerial coverage, topicality, availability, costs of acquisition, and data format. Among these factors, the map accuracy often only is a subordinate aspect. The user can only accept (or reject) the quality of a land-use/land-cover map as an a priori characteristic of a mapping product. Assessing the accuracy of large-area land-use/land-cover maps is problematic because the reference data to conduct the assessment are difficult and expensive to obtain. Therefore, time and budget constraints of research projects normally eliminate any possibility to verify the accuracy of a digital land-use/land-cover map used as model input.

For the past nine years, the land-use research project (SFB 299 “Land use options for peripheral regions”, details refer Frede et al., 2002) has analyzed and evaluated the multifunctionality of landscapes from different perspectives. The core of the SFB 299 project is an integrated modeling approach that includes models dealing with floristic diversity (Waldhardt et al., 2004), faunistic diversity (Dauber et al., 2003), agronomy (Kuhlmann et al., 2002), and hydrology (Huisman et al., 2004). Within this research project, there was the need to select the most appropriate land-use map to be used as input data for the integrated modeling.

In most European countries, several digital land-use mapping products are available: national digital topographic maps (typically 1:10,000 or 1:100,000), the digital land-use data set of the EU project “CORINE Land Cover” (scale 1:100,000), and remote sensing products such as land-use classifications from satellite images (for a detailed overview see Meinel and Hennersdorf, 2002). In the case of the SFB 299 project region, three digital land-use/land-cover maps are available:

  • the digital topographic map of Germany (ATKIS),

  • the CORINE Land Cover (CLC, coordination of information on the environment) data set for Germany,

  • a Landsat 5 TM classification from Noehles (2000).

For these three maps, two issues were of concern: (i) an a posteriori assessment of the accuracy of the three land-use data sets by comparison with the results of a field inventory in two smaller test areas, and (ii) an assessment of the possibility to substitute one map by another by testing the congruency of the geo-objects of homologous land-use classes for the three land-use data sets for the entire study region. The results for the larger region are also compared to the statistical land-use data in the cadastral register. In a strict sense, the results of both the accuracy and the congruency assessment are valid only for the mapping area of the study region. However, a discussion of the divergences between the digital maps (and the associated uncertainty) is relevant for a broad range of map applications.

Section snippets

Study region

The study region corresponds to the project area of the joint research center SFB 299 (Frede et al., 2002) and comprises 14 municipalities with an area of 670 km2 (Fig. 1) covering approximately the drainage basin of the river Dill. The area of interest is located in the northern part of the Lahn-Dill county in the mid-Hessian highlands and is characterized by soil and climate conditions that are relatively unfavorable for agricultural production. Furthermore, the farmland has typically very

Accuracy test of the land-use maps

The land use according to the field inventory and the three digital land-use maps for the two test areas are compared in Table 2. For urban and traffic areas, the largest absolute difference was 1.4% in the district Erda and around 3% in the district Stb./Eibh. In the case of forest, the field inventory in Erda resulted in 37.3%, while the land-use maps varied between 33.8% and 41.9%. The largest difference between two land-use data sets for one of the four main land-use classes is 8%. In

Discussion

When all the six land-use classes are considered, the digital land-use maps ATKIS, CLC, and Landsat 5 TM classification by Noehles (2000) represent the actual land-use distribution with an overall accuracy of 0.876 at best (ATKIS for the test area of district Erda). However, the accuracy can drop as low as 0.7, for example, in the case of the CLC dataset for Stb./Eibh. When the focus is on an analysis of individual land-use classes of the open land (arable land, pastures and meadows, and fallow

Conclusions

The widespread application of spatially distributed model approaches has led to an increased demand for thematic maps depicting land use and land cover. Because of inadequate funding, time, or training these maps are produced with little consideration given to the quantification and documentation of their accuracy (Dicks and Lo, 1990). Every model application based on land-use/land-cover data requires certain knowledge about their quality or at least an idea of the possible uncertainty of

Acknowledgements

We thank the Hessian Land Survey Office for the provision of the ATKIS data. The Joint Research Project SFB 299 is financially supported by the German National Research Foundation, Bonn. The very useful comments of the two reviewers significantly improved the presentation and the focus of the article.

Martin Bach is research associate at the Institute of Landscape Ecology and Resources Management, Division of Natural Resources Management at the University Giessen. Since 1997 he is in charge as the scientific coordinator of the DFG-funded joint research project (SFB) 299 “Land Use Options for Peripheral Regions”. His current fields of research are the analysis and modelling of pesticide and nutrient fluxes at different scales, and multifunctionality of landscape functions.

References (47)

  • S.V. Stehman

    Selecting and interpreting measures of thematic classification accuracy

    Remote Sens. Environ.

    (1997)
  • S.V. Stehman et al.

    Design and analysis for thematic map accuracy assessment: fundamental principles

    Remote Sens. Environ.

    (1998)
  • V. Walter

    Object-based classification of remote sensing data for change detection

    ISPRS J. Photogramm. Remote Sens.

    (2004)
  • AdV, 2002. Amtliches topographisch-kartographisches Informationssystem (ATKIS) – Gesamtdokumentation....
  • BKG, 2000. Vektordaten Verwaltungsgrenzen (Gemeinden) 1:250 000 (VG250). Bundesamt fuer Kartografie und Geodaesie,...
  • P.V. Bolstad et al.

    Errors in GIS: assessing spatial data accuracy

    J. Forest.

    (1992)
  • A. Busch et al.

    Automated verification of a topographic reference dataset: system design and practical results

  • Commission of the European Communities, 1992. EUR 12585 – CORINE Land Cover Project – Technical Guide. Office for...
  • R.G. Congalton et al.

    Assessing the Accuracy of Remotely Sensed Data: Principals and Practices

    (1999)
  • R. Costanza

    Model goodness of fit: a multiple resolution procedure

    Ecol. Model.

    (1989)
  • S.E. Dicks et al.

    Evaluation of thematic map accuracy in a land-use and land-cover mapping program

    Photogramm. Eng. Remote Sens.

    (1990)
  • FAO, 2003. Overview of the Africover initiative. Food and Agriculture Organisation of the UN, 6 pp. Available at...
  • Federal Office of Statistics, 1996. Datenerhebungsanleitung CORINE Land Cover (deutsche Ausgabe), Stand 1.9.1996....
  • Cited by (64)

    • Revealing the transversal continuum of natural landscapes in coastal zones - Case of the Turkish Mediterranean coast

      2018, Ocean and Coastal Management
      Citation Excerpt :

      As there is a variety of LULC mapping techniques, the correct selection criteria are necessary in order to allow the use of the most appropriate. According to Bach et al. (2006) the selection criteria are dependent on a range of aspects, such as; thematic content, scale, spatial resolution, classification system, aerial coverage, topicality, availability, costs of acquisition, and the data format of the LULC map. Consequently, within the scope of this study CLC data was utilized for the following advantages: it is open source via European Environment Information and Observation Network (EIONET) of EEA-thereby nullifying the acquisition costs; it has a spatial resolution appropriate for the eco-regional scale of the study area; it has a clear thematic content; it provides a detailed classification system; and it is available in the shapefile format, making it appropriate for analytical processes via GIS technology.

    • Trajectory analysis of land use and land cover maps to improve spatial–temporal patterns, and impact assessment on groundwater recharge

      2017, Journal of Hydrology
      Citation Excerpt :

      To assess the effects of land use change on groundwater recharge in Flanders, three CORINE land cover maps (CLC, 1990, 2000, 2006) were resampled from 100 m to 50 m resolution using nearest neighbor method in ARCGIS. The maps include 44 classes of LULC features, which were primarily classified by visual image interpretation rather than automated classification procedures (Bach et al., 2006). The land use map for Flanders of 2013 (Poelmans et al., 2014) has a resolution of 10 by 10 meter, its legend includes 114 detailed land use classes, which were structured hierarchically.

    View all citing articles on Scopus

    Martin Bach is research associate at the Institute of Landscape Ecology and Resources Management, Division of Natural Resources Management at the University Giessen. Since 1997 he is in charge as the scientific coordinator of the DFG-funded joint research project (SFB) 299 “Land Use Options for Peripheral Regions”. His current fields of research are the analysis and modelling of pesticide and nutrient fluxes at different scales, and multifunctionality of landscape functions.

    Lutz Breuer is a research associate the Division of Natural Resources Management. He took his PhD in Forestry at the University of Freiburg (Germany), where he investigated N-trace gas emissions from tropical rainforest soils. His focus in research is now on hydrological and biogeochemical modelling on the landscape scale. He has a strong interest in land use change studies and integrated approaches to analyse the multifunctional effects of global change.

    Hans-Georg Frede is Professor and Head of Division of Natural Resources Management. At present he is the Speaker of the joint research project (SFB) 299 and other research projects on the topic of simulation of water flow in vegetated filter strips, mapping of satellite scenes, surface sealing by erosion, modelling pesticide transport and calculating balance sheets for large scales. He was organizer of several international congresses and workshops concerning pesticide pollution of flowing waters. Furthermore he participates in the EU research project “Soil and water quality as affected by agrochemicals”.

    J.A. (Sander) Huisman received a master's degree (1997) with honors from the department of Physical Geography and Soil Science, Universiteit van Amsterdam, Netherlands. In 2002, he finished his PhD at the Universiteit van Amsterdam on the use of ground-penetrating radar (GPR) as a soil water content sensor. From 2002 to 2005, he worked as a postdoctoral researcher in SFB 299, University of Giessen. Currently, he is working at the Agrosphere Institute of the Forschungszentrum Jülich (Germany). His main research interests are hydrogeophysics, vadose zone hydrology and meso-scale hydrological modeling.

    Annette Otte is Professor and Head of the Division of Landscape Ecology and Landscape Planning at Giessen University. She has long time experiences in research and teaching in the fields of analysis of biodiversity in agricultural landscapes and biodiversity management and she is in a leading position in numerous scientific societies, journal editorial boards and advisory boards of scientific organisations.

    Rainer Waldhardt received 1994 his PhD-diplom at the Faculty of Natural Sciences, University Goettingen, Germany. 1995 he started as postdoctoral research associate at the Division of Landscape Ecology and Landscape Planning, tackling the spatio-temporal scales in vegetation and landscape ecology as fundamental research issues. Since 2003 he is second project manager of the subproject ‘Modules for the modelling of land-use dependent plant species diversity in cultural landscapes’ of the SFB 299.

    View full text