EEA Technical report No 7/2006 The thematic accuracy of Corine land cover 2000 Assessment using LUCAS (land use/cover area frame statistical survey) ISSN 1725-2237 X EEA Technical report No 7/2006 The thematic accuracy of Corine land cover 2000 Assessment using LUCAS (land use/cover area frame statistical survey) Cover design: EEA Layout: EEA Legal notice The contents of this publication do not necessarily reflect the official opinions of the European Commission or other institutions of the European Communities. Neither the European Environment Agency nor any person or company acting on behalf of the Agency is responsible for the use that may be made of the information contained in this report. All rights reserved No part of this publication may be reproduced in any form or by any means electronic or mechanical, including photocopying, recording or by any information storage retrieval system, without the permission in writing from the copyright holder. For translation or reproduction rights please contact EEA (address information below). Information about the European Union is available on the Internet. It can be accessed through the Europa server (www.europa.eu). ISBN 92-9167-844-9 ISSN 1725-2237 © EEA, Copenhagen, 2006 European Environment Agency Kongens Nytorv 6 1050 Copenhagen K Denmark Tel.: +45 33 36 71 00 Fax: +45 33 36 71 99 Web: www.eea.europa.eu Enquiries: www.eea.europa.eu/enquiries Contents Contents Acknowledgements ....................................................................................................4 Executive summary ....................................................................................................5 1 Introduction .........................................................................................................7 1.1 Background ....................................................................................................7 1.2 Purpose .........................................................................................................8 2 Databases ............................................................................................................9 2.1 CLC2000 data .................................................................................................9 2.2 Image2000 data .............................................................................................9 2.3 LUCAS 2001–2002 data ...................................................................................9 2.4 Setting up the database ...................................................................................9 2.5 CLC2000 cross-tabulated with LUCAS ................................................................11 2.6 Representativeness ........................................................................................15 3 Methodology ......................................................................................................18 3.1 Automatic comparison ...................................................................................18 3.2 Evaluation based on reinterpretation of LUCAS photographs .................................19 4 Results of the automatic approach .....................................................................21 5 Results of the reinterpretation ...........................................................................24 5.1 Reliability of CLC2000 .....................................................................................24 5.2 Analysis of errors ...........................................................................................26 5.3 Subjectivity of photo interpretation ...................................................................28 6 Remarks on LUCAS data quality .........................................................................29 7 Conclusions ........................................................................................................30 References ...............................................................................................................32 Annexes overview ....................................................................................................33 Annex 1 Corine land cover nomenclature ................................................................34 Annex 2 LUCAS land cover nomenclature ................................................................35 Annex 3 LUCAS land use nomenclature ...................................................................36 Annex 4 Results of cross-tabulating CLC2000 and LUCAS LC ..................................37 Annex 5 Results of cross-tabulating CLC2000 and LUCAS LU ..................................38 Annex 6 Correspondence table between CLC classes and LUCAS land cover and land use classes .................................................................................39 Annex 7 Reclassification of Image2000 using LUCAS: error matrix.........................40 Annex 8 Reclassification of Image2000 using LUCAS: subjective matrix .......................................................................................................41 Annex 9 Illustrations (screen shots taken during reinterpretation — see Section 3.2) ....................................................................................42 The thematic accuracy of Corine land cover 2000 3 Acknowledgements Acknowledgements The CLC2000 project is part of the work programme and support. Thanks also to Vanda Lima and of the European Topic Centre on Terrestrial Javier Gallego (JRC) for the useful comments on Environment (ETC-TE), working under contract the validation methodology. Special thanks to with the European Environment Agency Mária Bíró, who performed the reinterpretations (2001–2005). LUCAS data for the study were and László Oszkó for his help on statistical provided by Eurostat and the national statistical considerations. Linguistic corrections were provided offices in Estonia, Hungary and Slovenia. The by Barbara Kosztra. authors express their gratitude to Chris Steenmans, EEA project manager, Adriana Gheorghe, Prepared by George Büttner and Gergely Maucha, EEA-Eionet coordinator, and Stefan Kleeschulte, ETC/TE partner FÖMI. ETC-TE manager for continuous cooperation 4 The thematic accuracy of Corine land cover 2000 Executive summary Executive summary Corine land cover 2000 (CLC2000) was produced The specific questions that the validation procedure by the European Environment Agency (EEA) and is supposed to answer are: its member countries in the European environment information and observation network (Eionet), • Was the 85 % thematic accuracy target of based on the results of IMAGE2000, a satellite CLC2000 fulfilled? imaging programme undertaken jointly by the Joint • Which classes were determined with high Research Centre of the European Commission and accuracy and which ones with low? the EEA. CLC2000 is an update for the reference • What were the reasons for misinterpretations? year 2000 of the first Corine land cover database • How subjective was photo-interpretation? which was finalised in the early 1990s as part of the European Commission programme to COoRdinate Two kinds of methods were used for validating Information on the Environment (Corine). The CLC2000 data: resulting national land cover inventories were integrated into a seamless land cover map of • Reinterpretation of IMAGE2000 data based on Europe, which is based on a standard methodology LUCAS codes and photographs; and nomenclature, providing consistent information • Automatic comparison of CLC2000 codes and on land cover across Europe. LUCAS LU and LC codes. LUCAS (European Land Use/Cover Area Frame The main result of the reinterpretation approach is Statistical Survey) is a project managed by the that the total reliability of CLC2000 is 87.0 ± 0.8 %, Eurostat, which main purpose is to provide which leads us to the conclusion that the 85 % harmonised information on agri-environment for accuracy requirement specified in the Technical Europe. LUCAS records land use (LU) and land Guidelines of CLC2000 was fulfilled. cover (LC) information and stores digital field photographs in a two-level regular grid. The result of the automatic comparison was that the percentage of total agreement (PTA) found between The purpose of this report is to assess the thematic CLC2000 and LUCAS LU and LC is 74.8 ± 0.6 %, accuracy of CLC2000 by means of an independent meaning that CLC2000 approximates LUCAS data source, the LUCAS data (not used in the thematic data with a 74.8 % average accuracy. compilation of the target database — CLC2000) and statistical methods homogeneously applied to The analysis of results at class level shows that only as many of the participating countries as possible. 22 of the 44 classes of CLC2000 could be validated. The purpose of validating geographical data is to The 22 classes that could not be validated belonged derive final accuracy/reliability figures by means mostly to artificial surfaces, wetlands and water. of independent, high resolution and more accurate data, which is related to a similar data acquisition The highest class-level reliability (> 95 %) was period. obtained for rivers, lakes, industrial and commercial units and discontinuous urban fabric. LUCAS data is the only information that is available in many countries (18) for a European- The two largest CLC classes (arable land and wide validation of CLC2000. LUCAS fulfils the coniferous forest) were estimated to have a high criteria of validation data as a result of its following level of reliability (between 90 and 95 %). Two other characteristics: agricultural classes also enjoyed a high level of reliability: agro-forestry and permanently irrigated • high geometric accuracy; land. • data acquisition nearly coincident with CLC2000; The lowest class-level reliability (below 70 %) was • independent, as not used during production of obtained for the sparse vegetation class, which CLC2000. highlights the difficulties in interpreting this category. The thematic accuracy of Corine land cover 2000 5 Executive summary Main sources of mistakes included using the wrong The validation of CLC2000 leads us to think of some code (commission errors) and not interpreting potential applications of the results. In the next enough details (omission errors). Geometrical update of CLC, special attention should be paid to inaccuracies played a less important role. the less accurate classes which mean that there is a need for improvement of the definition of mapping The majority of classification errors (78 %) occurred rules and the use of multi-temporal satellite data at level-3 and level-2 of the classification. Level-1 during interpretation. Also, positive examples of misclassifications mostly occur between the simultaneous display of IMAGE2000, CLC2000 and agriculture and the forest and semi-natural classes. LUCAS data could be used in CLC related training/ demonstration activities. And finally, the European The analysis revealed that subjectivity of photo- LUCAS survey in 2006 could benefit from the results interpretation could be noticed in 18 % of the by improving consistency of LU/LC codification and samples. The most subjective CLC classes are, as producing better field photographs. follows: agriculture with significant amount of natural vegetation, transitional woodland, shrub, complex cultivation patterns and mixed forest. 6 The thematic accuracy of Corine land cover 2000 Introduction 1 Introduction 1.1 Background first CLC inventory (revised CLC90) and metadata (EEA-ETC/TE, 2002). Corine land cover 2000 (CLC2000) is a project jointly managed by the Joint Research Centre (JRC) and the European land use/cover area frame statistical European Environment Agency (EEA). Its aim is to survey (LUCAS) is a project managed by Eurostat. update the Corine land cover database in Europe Its main purpose is to provide harmonised for the year 2000. Landsat-7 Enhanced Thematic information on agri-environment for Europe. Mapper (ETM) satellite images were used for the LUCAS (as implemented in 2001–2003) recorded update and were acquired within the framework of land use and land cover information in a two- the Image2000 project. The first Corine land cover level regular grid. The grid size was 18 x 18 km (CLC) inventory for the EU-15 and most of the new (primary sampling unit, PSU). Each PSU included Member States was implemented between 1985 and 10 secondary sampling units (SSUs). SSUs were 1996. It was carried out in order to characterise the placed in two parallel rows, with five points in each land surface. A uniform nomenclature across Europe row. The distance between SSU points was 300 m at scale 1:100 000 was used. The CLC nomenclature (Figure 1). The field surveyor located the point with basically includes land cover items though land a high level of accuracy (specification states 1–3 m), use elements can also be found. This is especially and registered LC and LU information related to the case for built-up environments (Heymann et al., this point. This was conducted according to strict 1994; Bossard et al., 2000). The CLC database is a guidelines (Avikainen et al., 2003). The area of digital map covering countries with a seamless observation was usually a circle with a 1.5 m radius. polygon database with 25 ha minimum mapping However, in exceptional cases an area with a 20 m unit (MMU). The CLC nomenclature includes radius was considered (Duhamel et al., 2003). 44 categories in five major groups (Annex 1). In addition to the updated CLC2000 there are also Usually, in each central SSU point of the first row other national deliverables: land cover changes (SSU13), digital field photographs were taken in between the 1990s and 2000 (CLC-Change), revised east, west, north and southwards directions. Figure 1 The LUCAS two-stage sampling (LUCAS technical document No 1: sampling plan) The thematic accuracy of Corine land cover 2000 7 Introduction The LUCAS LC nomenclature includes 57 categories • high geometric accuracy; in seven major groups (Annex 2), while LUCAS LU codification applies 14 land use classes (Annex 3). • high thematic accuracy (LC/LU data based on field checking or, in exceptional cases, on photo A new LUCAS survey with a denser grid will take interpretation of aerial ortho-photographs); place in 2006 in 23 Member States (Eurostat, 2005). • data acquisition nearly coincide (although in the worst case a three-year difference is possible); 1.2 Purpose • independent, as usually not used during The purpose of validating geographical data is to production of CLC2000. derive final accuracy/reliability figures by means of independent, high resolution and more accurate data, which are related to a similar data acquisition The specific questions of the validation are the period. Furthermore, they were not used in the following. compilation of the target database. Some countries performed a national validation of their CLC2000 • Is the 85 % target accuracy (Corine land database using country-specific approaches. In most cover update, I&CLC2000 project, Technical cases, more spatially and thematically detailed data guidelines, Final version, August 2002) fulfilled? were used for that purpose (for example, see final reports in the Netherlands, Italy, Finland, Sweden, • Which classes are determined with high Portugal and the United Kingdom). Validation accuracy and which ones with low accuracy? in this context means thematic validation as the geometry of CLC2000 is too trivial to check by • What are the reasons for misinterpretations? comparison with Image2000. • How subjective is photo interpretation? The general aim of validating CLC2000 is to assess its thematic accuracy by means of a statistical method which is homogeneously applied to as many Due to lack of similar reference data with wide of the participating countries as possible. coverage for the period of the first CLC inventory (1985–1996), the two other deliverables of the LUCAS data are the only information available in CLC2000 project (CLC-Change and revised CLC90) many countries for a European-wide validation of are not validated. CLC2000. LUCAS fulfils the criteria of validation data as a result of its following characteristics: The results of the validation are made available for the users of CLC2000, and could be especially valuable in planning the next update of Corine land cover. 8 The thematic accuracy of Corine land cover 2000