International Journal of Applied Earth Observation and Geoinformation
Changes in the methodology used in the production of the Spanish CORINE: Uncertainty analysis of the new maps
Introduction
Land Use Land Cover (LULC) maps are extremely useful data, as evidenced by their widespread use in a large number of disciplines, such as Geography. These maps provide information on land occupation, seen as the biophysical cover observed on the Earth’s surface (Delgado Hernández, 2008). Their use is essential for the study and modeling of territorial dynamics.
There are different sources on land use and land cover for the Spanish territory. At global and continental levels maps are generated from supervised or unsupervised classification techniques. Some of the most significant at a spatial resolution of 1 km are GLC2000, UMD Land Cover Classification and DisCover. The first one was created for the year 2000, and the rest for the period 1992/93. At a finer resolution we can find MODISLC (500 m), drawn up yearly since 2000, and GlobCover (300 m), with reference date 2005.
For Europe, PELCOM was created from images of the year 1996, with a resolution of 1 km. At the national and regional level, CORINE (Coordination of information on the environment) and SIOSE (Land Cover and Land Use Information System of Spain) maps stand out, obtained from photointerpretation.
Comparative studies of different LULC maps at the global or continental level (Fritz and See, 2005; Giri et al., 2005; Hansen and Reed, 2000; Herold et al., 2008; McCallum et al., 2006; Tchuenté et al., 2011), some of which have used CORINE (Neumann et al., 2007), have evidenced differences and contradictions between the different sources. There are also many studies at the national and regional levels with similar conclusions (Bach et al., 2006; Chas-amil and Touza, 2015; Pérez-Hoyos and García-Haro, 2009; Ran et al., 2010; Waser and Schwarz, 2006).
In a similar vein, several studies suggest that the accuracy of these maps must be assessed. For CORINE, there are several analyses about the errors and confusions of its database (Caetano et al., 2006; Catalá Mateo et al., 2008; Diaz-Pacheco and Gutiérrez, 2013; Teixeira et al., 2016), although the uncertainties and limitations of these maps have not been totally specified.
Hence, there is a strong uncertainty linked to the way in which each map includes the information about land use and land cover. Understanding this uncertainty is essential for all research based in these maps, as in the case of LULC modeling (García-Álvarez in press; Hewitt et al., 2014, Camacho Olmedo et al., 2015). It is therefore necessary to assess the different ways used to produce CORINE in different countries and sometimes even within a single country.
Traditionally, CORINE has been generated from the photointerpretation of satellite imagery. However, in some countries (Germany, Austria, Finland, Ireland, Iceland, Norway, the United Kingdom, Sweden, Switzerland), especially since CLC06, the map is obtained from generalization techniques of national maps of greater detail (Hazeu et al., 2016). In other cases (Slovakia, Hungary, Poland, etc.), the CORINE map is used to obtain other maps of greater detail, at a 1:50.000 scale, with a minimum mapping unit (MMU) of 4 ha and a legend adapted to the geographical specifications of the country (Hazeu et al., 2016). They have even been used to obtain CORINE maps from before 1990 (Feranec et al., 2007a). In the case of Spain, CLC90 and CLC00 have a more detailed legend with 85 classes, with exceptions in the MMU (25 ha) for water surfaces and artificial areas (Catalá Mateo et al., 2008).
Therefore, there are different ways of producing CORINE, each of which can generate a different result. In Germany and Ireland, the methodological change in the production of CORINE (from photointerpretation to generalization) has led to the production of new maps, which show important inconsistencies with respect to the previous ones (Hovenbitzer et al., 2014, Lydon and Smith, 2014). In the Netherlands, this limitation has been one of the arguments for continuing to produce CORINE independently (Büttner, 2014; Hazeu et al., 2016).
The recent change in the methodology of production of CORINE in Spain (Hazeu et al., 2016) must go hand in hand with studies which show the degree of accuracy of the new CORINE map and its differences with the previous ones. Since the year 2012 CORINE has been obtained by generalizing SIOSE, as opposed to the digital photointerpretation technique used until the year 2006. For the same date CORINE was obtained using both methods (photointerpretation and generalization). This allows us to compare the CORINE obtained through both methods for the same time and place.
This paper aims to compare CORINE 2006 for Spain, obtained according to both the methods described above. With this we aim to answer the following questions: How has the change in production methods affected the information supplied by CORINE? Can the new CORINE be compared to the previous one? Is it more precise and reliable? How has the generalization from SIOSE to CORINE been performed and what degree of uncertainty is conveyed?
With these goals in mind, this paper is structured as follows: first, we begin by providing brief introduction to the CORINE and SIOSE databases, showing their features, differences and, in the former case, the recent change in its production methodology. We then describe the study area and the tools with which the analysis has been carried out. Finally, the results are presented and discussed in relation to the problems with the production of CORINE Spain and the rest of Europe.
Section snippets
CORINE
CORINE is a LULC map developed initially in 1990 (CLC90), with updates for the years 2000 (CLC00), 2006 (CLC06) and 2012 (CLC12). Nowadays, it is updated every six years and comprises most of the European continent, although in each update an increasing number of countries have been taking part in the initiative (Büttner, 2014).
CORINE is produced following a hierarchical data model or classification system. A set of classes (legend) is defined and organized in a hierarchical manner. Each
Study area
The chosen study area is the Asturias Central Area (ACA), as delimited in the Asturias Regional Guidelines for Land Planning (Fig. 3). It comprises the area of greatest demographic and economic development in Asturias (Fernández García et al., 2007), as well as most of the territorial changes in the region.
The Asturian landscape, characterized by overlapping and fragmentation of different LULCs, is very heterogeneous. The rural land shows an excessive segmentation (Dirección General de
Method
A cross tabulation of the two CORINE maps for 2006, one created using digital photointerpretation (CLC06t) and the other by the generalization of SIOSE (CLC06g) (Fig. 4), has been carried out using the matrix designed by Pontius and Millones (2011), available at http://www2.clarku.edu/∼rpontius/. This enabled us to find out the quantity and allocation disagreement of the classes.
Quantity disagreement refers to the amount of difference between the quantities or proportions of every class in two
Differences in the quantity and allocation of CORINE classes
Only half of the surface of both CORINE maps for the year 2006 shows the same information (percentage of total difference of 50%) (Fig. 7). In the disagreement, 23% corresponds to a quantity disagreement, i.e., certain classes are overrepresented in a map in relation to the other, while others are underrepresented. Within the allocation disagreement, 10% is considered exchange (pair wise confusions), while the remaining 17% corresponds to shift (non-pair wise confusions).
The quantity
Uncertainties linked to generalization methodologies
The polygons delimited by the new CORINE (CLC06g) present more complex and detailed perimeters than those of earlier CORINE maps obtained by photointerpretation (CLC06t). These are the perimeters sketched at the SIOSE photointerpretation scale (1:25,000) which therefore do not correspond to the lower degree of complexity expected at a 1:100,000 scale. A possible solution to this problem would be to implement geometrical simplification mechanisms, as in other countries (Brown et al., 2002; Härmä
Conclusions
The change of methodology in the production of CORINE in Spain has generated a new map that is different from the previous ones. Obtaining it from the generalization of SIOSE has improved the consistency between the different LULC maps for Spain and savings have been achieved in the production costs. However, the new CORINE presents important uncertainty sources.
Obtaining it from a process of quantitative generalization cannot be compared with the task traditionally carried out by
Acknowledgements
This work has been supported in part by project SIGEOMOD_2020·BIA2013-43462-P (Spanish Ministry of Economy and Competitiveness and the Feder European Regional Development Fund). The lead author is also grateful to the Spanish Ministry of Economy and Competitiveness and the European Social Fund for the funding of his research activity (Ayudas para contratos predoctorales para la formación de doctores 2014).
References (73)
- et al.
Accuracy and congruency of three different digital land-use maps
Landsc. Urban Plan.
(2006) - et al.
Comparison of simulation models in terms of quantity and allocation of land change
Environ. Model. Softw.
(2015) - et al.
Inventory of major landscape changes in the Czech Republic, Hungary, Romania and Slovak Republic 1970–1990s
Int. J. Appl. Earth Obs. Geoinf.
(2000) - et al.
Corine land cover change detection in Europe (case studies of the Netherlands and Slovakia)
Land Use Policy
(2007) - et al.
Corine land cover change detection in Europe (case studies of the Netherlands and Slovakia)
Land Use Policy
(2007) - et al.
Determining changes and flows in European landscapes 1990–2000 using CORINE land cover data
Appl. Geogr.
(2010) Interpretation element association: analysis and definition
Int. J. Appl. Earth Obs. Geoinf.
(1999)- et al.
Using CORINE land cover and the point survey LUCAS for area estimation
Int. J. Appl. Earth Obs. Geoinf.
(2008) - et al.
Using remote sensing products to classify landscape. A multi-spatial resolution approach
Int. J. Appl. Earth Obs. Geoinf.
(2016) - et al.
A comparative analysis of the Global Land Cover 2000 and MODIS land cover data sets
Remote Sens. Environ.
(2005)
Finding the right pixel size
Comput. Geosci.
Some challenges in global land cover mapping: an assessment of agreement and accuracy in existing 1 km datasets
Remote Sens. Environ.
The territorial dynamics of fast-growing regions: unsustainable land use change and future policy challenges in Madrid, Spain
Appl. Geogr.
Participatory land use modelling, pathways to an integrated approach
Environ. Model. Softw.
A spatial comparison of four satellite derived 1 km global land cover datasets
Int. J. Appl. Earth Obs. Geoinf.
Comparative assessment of CORINE2000 and GLC2000: Spatial analysis of land cover data for Europe
Int. J. Appl. Earth Obs. Geoinf.
Resolution dependent errors in remote sensing of cultivated areas
Remote Sens. Environ.
Conventional and fuzzy comparisons of large scale land cover products: application to CORINE, GLC2000, MODIS and GlobCover in Europe
ISPRS J. Photogramm. Remote Sens.
DINAMICA − a stochastic cellular automata model designed to simulate the landscape dynamics in an Amazonian colonization frontier
Ecol. Modell.
The effect of field experience on the accuracy of identifying land cover types in aerial photographs
Int. J. Appl. Earth Obs. Geoinf.
Evidence for deviations from uniform changes in a Portuguese watershed illustrated by CORINE maps: an intensity analysis approach
Ecol. Indic.
Comparison of large-area land cover products with national forest inventories and CORINE land cover in the European Alps
Int. J. Appl. Earth Obs. Geoinf.
Corine Land Cover Classes. Examination of the content of CLC classes in Norway
CORINE Land Cover and Land Cover Change Products
Mapping of land use classes within the CORINE Land Cover Map of Great Britain
Cartogr. J.
Accuracy assessment of the Portuguese CORINE Land Cover map
Elaboración y validación de un modelo jerárquico derivado de SIOSE
Rev. Teledetec.
Understanding the scale and resolution effects in remote sensing and GIS
Análisis de posibles errores en la base de datos Corine Land Cover (1990–2000) en la Comunidad de Madrid
Estud. geográficos
Assessment of the Spanish land cover information to estimate forest area in Galicia
Boletín la Asoc. Geógrafos Españoles
Exploring and Evaluating the Consequences of Vector-to-Raster and Raster-to-Vector Conversion
Photogramm. Eng. Remote Sens.
Asturias
Application of CORINE land cover methodology to the U.K.-some issues raised from northern Ireland
Glob. Ecol. Biogeogr. Lett.
Armonización de las bases de datos de ocupación del suelo y su importancia en la evaluación de parámetros e indicadores medioambientales
Methodology of classification extraction from descriptive systems of land cover and land use
Cited by (43)
Combining multiple geostatistical analyses to assess the past, present, and future of fragile Mediterranean deltaic environments
2023, Journal of Geochemical ExplorationDealing with the uncertainty of technical changes in the CORINE Land Cover dataset: The Portuguese approach
2023, International Journal of Applied Earth Observation and GeoinformationEvaluation of fire severity in fire prone-ecosystems of Spain under two different environmental conditions
2020, Journal of Environmental ManagementCitation Excerpt :This database (year 2014, Spanish coverage, 1:25000 scale) (Ministerio de Fomento, 2014) is derived from orthophotography, satellite imagery and field data, with a minimum mapping unit of 2 ha. It provides information on the proportion of cover considering single land use classes consisting of a single element (e.g., conifer), or complex classes resulting from the association of several elements (e.g., regular mosaic: 90% olive trees and 10% shrubland; García-Álvarez and Camacho-Olmedo, 2017). Based on the SIOSE database, we classified vegetation types within the fire perimeters into four categories including forests (i.e., broad-leaf forest and conifer for the Mediterranean-Oceanic transition and Mediterranean sites, respectively), fruit tree crops (i.e., olive and carob tree crops, present in the Mediterranean site only), shrubs and grasslands (i.e., for the Mediterranean-Oceanic transition site only).
Sensitivity of a common Land Use Cover Change (LUCC) model to the Minimum Mapping Unit (MMU) and Minimum Mapping Width (MMW) of input maps
2019, Computers, Environment and Urban SystemsCitation Excerpt :Although testing the sensitivity of a model to different datasets with different MMU and MMW would be closer to real practice, it would prevent us from achieving our objective, in that the differences between the datasets would not only be due to scale, but also to the different methods of production, classifications, etc. In this regard, the only difference between our input maps was their scale, as CORINE was obtained from a generalization of SIOSE (García-Álvarez & Camacho Olmedo, 2017). The two input maps were rasterized at the same spatial resolution (50 m) so as to make sure that the rasterization process did not interfere with the analysis.