Validation of the CORINE land cover database in a pilot zone under semi-arid conditions in La Mancha (Spain)
Résumés
Le programme CORINE (qui débute en 1985) a créé un système d’information relatif aux pays de L’Union Européenne qui fournit des données administratives, géographiques, utilisation du sol, couverture végétale et aussi différents paramètres sur l’environnement (CORINE,1994). La base de données « CORINE land cover » offre une information précieuse sur les différentes occupations du sol dans le secteur cartographié. Actuellement, toute la Péninsule Ibérique est accessible dans cette base de données numérisée.
CORINE constitue un instrument de grande valeur dans les pays communautaires où elle est déjà finie. Il est pourtant nécessaire de valider la qualité de l’information qu’il offre, étant donné la grande extension et la variété climatique et environnementale de la superficie cartographiée.
Ce travail est une étude détaillée, en partant de sources mises à jour, de l’état de l’occupation du sol sur un secteur situé dans la zone centrale de La Mancha (Espagne) de 10.000 km2 approximativement. Il est ensuite comparé avec les données CORINE pour ce même secteur. Il s’agit donc d’effectuer une validation dans une zone pilote en utilisant des bases de données qui ont une définition plus fine que celle de CORINE, et observer tant son degré de mise à jour que l’adéquation des différentes catégories utilisées par CORINE pour décrire l’utilisation du sol et les écosystèmes actuels.
Le travail sur données géographiques implique le choix d’une échelle appropriée qui dépend de la nature de l’étude à réaliser. CORINE s’avère une bonne base de données pour la gestion de l’information de la couverture végétale sur de grandes échelles de travail. Les erreurs de géoréférencement que nous détectons dans ce travail soulignent qu’il est très difficile de la superposer à d’autres bases de données.
Après avoir réalisé toutes les vérifications nécessaires, sans l’aide de la superposition, on a constaté que CORINE land cover commence à être dépassée, compte tenu des transformations, des changements et du développement de l’actuelle couverture végétale du sol dans notre région.
D’autre part, à notre avis, la nomenclature de CORINE land cover n’est pas assez diversifiée au troisième niveau, surtout dans les zones forestières. Cela fait que dans une même catégorie se groupent des formations bien distinctes entre elles.
Cela n’arrive pas dans les zones urbaines où il y a une grande quantité de catégories pour définir la variété d’utilisation du sol urbain. Cette étude a été réalisée dans le cadre du projet « Modelling the effect of land degradation on climate » (ENVA-CT98-0696) financé par l’ Union Européenne
Entrées d’index
Mots-clés :
CORINE land cover, instruments de planification de l’environnement, occupation du sol, SIG, validationPalabras claves :
instrumentos de planificación ambiental, sistemas de información geográfica, CORINE land cover, cobertura de suelo, validaciónPlan
Haut de pageTexte intégral
Introduction
1One of the most powerful tools that can be used in those projects which require some kind of assessment or management of the environment are the geographical Information Systems (GIS) . These systems enable us to build up geo-referenced databases capable of gathering a significant amount of information.
2On the other hand, the advances of remote-sensing and image-processing digitisation and analysis techniques contribute new programs and functions to the treatment, handling and development of maps which bring physical reality closer to laboratory work, thus causing the decision-making process to become increasingly based on reliable sources.
3It is therefore essential that these Geographical Information Systems should have the precision and accuracy which their growing use requires.
4The CORINE programme (started in 1985) has created an information system referred to the EU countries which provides data related to administration, geography, land uses and vegetal cover, as well as various environmental parameters (CORINE, 1994).
5The CORINE database on land cover provides valuable information on the various types of land cover in the area mapped. The whole Iberian Peninsula is covered at present by this digitised database. The information available through the database has been obtained from remote-sensing image-processing (Landsat MSS and TM, SPOT XS and P) as well as from statistical and vegetation maps. Its methodology was based on computer-aided photo-interpretation of this imagery contrasted with the various sources available on land cover. The resulting polygons were subsequently classified according to the CORINE nomenclature (European Topic Centre of Land Cover [on line], 2001).
6This nomenclature consists in 44 classes of land cover hierarchically arranged on three levels. The first level divides the surface area of the earth into artificial, agricultural and forestry areas, wetlands and continuous water surfaces. The last level details the types of land in the 44 classes mentioned above. This classification is based on a physical and physiognomic nomenclature related to land use, as well as on the vegetation parameters typical of each class.
7The inventory and digitising scale is 1:100,000 whereas the minimum surface area mapped is 25 hectares, the projection in which it is represented being of the Lambert Azymuthal type.
8This study has been carried out within the framework of the Project ‘Modelling the effect of land degradation on climate’ (ENV4-CT98-0696) funded by the EU. In this project, the research team of the University of Castilla-La Mancha have been responsible for providing the land cover data on the Iberian peninsula according to a classification compatible with the cover diagrams of global forecast climatic models.
9The CORINE database is rated as a tool of great value in the EU countries where it has been completed. It is therefore necessary to verify the quality of the information it provides taking into account the large size and the climatic and environmental variety of the surface area mapped.
10This paper consists of an in-depth study, based on various updated sources, of the actual state of the land cover in a specific area in the central part of Castilla-La Mancha (Spain) with a surface area of approximately 10,000 km2. The information obtained is then compared to the CORINE data for the same area. Our aim is therefore to verify the validity of the latter in a pilot zone by using, for this purpose, databases whose definition is higher than CORINE’s and observe both its degree of updated accuracy and the following up of the CORINE categories which appear in our zone and their evolution.
Materials and methods
11For the purpose of this study, a pilot zone was defined with a surface area of approximately 10,000 km2. This area, which has already been the object of several studies, most of which are within the framework of the projects EFEDA I (EPOC-CT-90-0030), EFEDA II (EV5V-CT 93-0272; 1993) and the CLD project mentioned above (ENV4-CT98-0696; 1998), is physically delimited by the coordinates 39:44:7.23N-2:53:24.26W; 39:34:31.64N-1:48:36.77W; 38:58:19.72N-3:7:26.27W and 38:48:48.24N-2:3:23.63W. Fig. 1 shows this study zone.
Figure 1 :Study zone. CORINE land cover database for the Iberian Peninsula. Close-up of the area.
12The first step was to gather the databases likely to be used in the comparison.
13The following sources have been used:
-
1:100,000-scale topographic map
-
Digitised Cadastral Survey of Rural Land.
-
Forest Map of Spain
-
Second National Forest inventory
-
Maps of Crop Classification by Tele-detection obtained from the University of Castilla-La Mancha.
14After analysing the above sources, we selected on a priority basis the Forest Map of Spain (MFE), the Maps of Crop Classification by Remote-Sensing (CCT) and the Digitised Cadastral Survey of Rural Land (CRD) as a base for geo-referencing purposes.
15The Topographic Map was used on various occasions to verify the situation of certain polygons and the only ‘classification of land use’ taken into account was that taken from the Second National Forest Inventory.
16The Forest Map of Spain (MFE) is a database on our country’s natural vegetation. The 1:50,000-scale digital version of the MFE has been implemented in the program ArcInfo and, attached to the geographical data, it contains the files which, through a numerical code relation, provide the meaning of each field. This is how we are given information on the surface, prevailing species, floristic aspect, subsoil, overload, presence of other species, inclusions, significant species, total vegetal cover, prevailing species cover, land use, etc. The minimum surface area mapped is 6,25 hectares, but it can be only 2,2 in isolated wooded areas between crops (ICONA, 1991). Its level of updated accuracy can still be deemed to be high as it was completed in 1995.
17As far as the Maps of Crop Classification by Remote-Sensing are concerned, this is a set of digitised maps generated by the study of satellite (Landsat) imagery based on the various values of the NDVI (Normalized Difference Vegetation Index) (Gilabert, 1996; Field and Bromdley, 1997) for each type of crop. This map has been developed and updated on a regular basis by the Remote-Sensing Section of the Regional Development Institute of the University of Castilla-La Mancha. The map used was updated in the year 2000, and it has a resolution of 25 m/pixel. (Calera et al., 2001; Martínez Beltrán et al.,2001)
18In order to be compatible the comparison, the database of all these maps was converted from its vectorial format into raster by using ArcInfo software.
19Moreover, CORINE also had to be converted from its original (Lambert Azymuthal) projection into the geographical coordinates in which the remaining databases are available.
20Comparisons were subsequently made at three levels by using the ArcView software intended for this purpose. At the first level, the geo-referencing quality of the various databases was verified by overlapping them with the Digitised Cadastral Survey of Rural Land. The maps were then compared at the second level by translating their various categories to those of the first hierarchical level of CORINE, i.e. artificial territories, agricultural lands, forestry and semi-natural areas, wetlands and water surfaces.
21Finally, we proceeded to the comparison at the third hierarchical level of CORINE. For this purpose, the nomenclature of MFE and CCT were translated, as with the previous level, to the CORINE categories. As far as the MFE was concerned, this translation was carried out for each polygon on a one-by-one basis by studying the various parameters involved and emphasizing the prevailing species as well as the characteristics of vegetal formation. So, the way in which CORINE land cover characterises each of its classes was taken into account to assign a CORINE category to each MFE Polygon.
22As to the CCT, because of the classification used based on spectral crop response, which was quite different from CORINE, the comparison was limited to three clustered classes as a result of the fact that the inclusion of a higher level of detail in the same class as CORINE proved to be impossible.
23The following classes were analysed: dry-land crops, which were compared to CORINE’s class 211, Irrigated-land Crops, which include CCT’s classes of alfalfa, irrigated spring and summer crops, as well as irrigated two-harvest fields, compared to CORINE’s class 212 and the vineyard class, which was compared to its counterpart in CORINE (Class 221).
24The comparison at this level was made in the form of percentages of surface area covered by each class, as CORINE’s deficient geo-referencing, which was found to be patent throughout our study, prevented us from overlapping the relevant polygons and calculating an agreement index. However, a comparison by overlapping was carried out from the beginning “a priori”, in case it did not provide any information, but as can be seen by the results, the percentages of coincidence forced us to reject this option.
Results
First level of comparison
25At the first level, the three databases (CORINE, MFE and CCT) were contrasted with the Digitised Cadastral Survey of Rural Land of the zone concerned in order to study how compatible these databases were when overlapped.
26The Cadastral Survey (CRD) was taken as a reference on a 1/50,000 scale of the study zone. This is a highly accurate and thoroughly updated source, with a resolution of 1 metre.
27Figures 2, 3 and 4 show an example of comparison over specific spaces of the three databases. The zone involved is an agricultural area in which the relevant polygons are perfectly defined by the CRD (boundaries between properties, roads and tracks, etc.) and the instances of non concordance between the different maps are evident. After observation in almost the whole study zone, the deficient delimitation of the polygons became conspicuous, particularly in CORINE, which makes it virtually impossible for the databases to overlap consistently. In several cases, deformation and displacement of the polygons can be detected in this data base. These deformations could originally be blamed on the CORINE land cover aggregation standards proposed in their Technical Guide, as also its lower resolution with respect to the CRD and two other databases, but there also exists the doubt whether there was really a geo-reference error, or whether the displacements and deformation are due to photo-interpreting errors in the genesis of CORINE.
28After this observation, we opted to compare the grid of UTM co-ordinates, scale 1/50.000 for the whole of the mainland, provided by the IGN (National Geographical Institute), with the respective grid 1/100.000 provided by the same CORINE database. It was here that we really proved the existence of a systematic displacement, in the whole of the mainland zone, which apart from displacing the polygons could cause the distortion of its limits. Our measurements gave us an average displacement of 1146 metres, a considerable distance taking into account the CORINE resolution, and that our intention was that of overlapping lower scale maps. An example of where the two grids overlap and the error is observed can be seen in Figure 5.
Figure 5: Outlet of Arc View for the zone of the Iberic mainland overlapping the grid 1/50.000 of the IGN with the grid 1/100.000 of CORINE ( zone shaded in green, grey line). The displacement observed is of about 1146 n between both.
Second level of comparison
29The comparison of the three databases using the first hierarchical level of CORINE gave the results shown in Figures 6 and 7.
30As it can be seen, the results show how the surface area of forestry increases as the updating level of the database becomes higher. This level is higher in the MFE than in CORINE, whereas in the CCT it is even higher than in the MFE. This data seems to be in agreement with the initial hypothesis of an increase in the surface area allocated to forestry, probably as a result of the re-foresting schemes carried out since the reform of the EC Common Agrarian Policy (PAC) of 1992.
31At any rate, it should be taken into account that the CCT takes the uncultivated areas of waste land surrounding cultivated zones as wooded territory which means that a small percentage of the total would correspond to these areas, where there is no forest use as such.
32A further point worth mentioning is that the MFE seems to have a lower percentage of the ‘water’ class owing to a deficiency in this database, i.e. the MFE does not record the mapping of rivers such as the Júcar and leaves out two major water surfaces like the reservoir of Peñarroya and the lagoons of Ruidera, in the South-West of the study zone, which do appear in the other two sources.
Figure 7: Percentage of classes at the first level in CORINE, MFE and CCT.
Third level of comparison
33At this level, as a first step intended for guidance, the CORINE data was divided into two major groups, the forestry classes and the agricultural classes compared to the MFE and CCT respectively.
34As a demonstration of the impossibility of overlapping the maps to see the equivalence of the polygons, we carried out an initial trail comparison as explained previously in the methods paragraph. In this ‘a priori’ comparison of surfaces indicates that, in the wooded zones, CORINE shows 47.6% coincidence with the CCT and 54.3% with the MFE (these percentages refer to the surface areas of forestry in CORINE that coincide with the surface areas of forestry of the other databases in the question of overlapping). In this way, the agricultural territories coincide to a greater extent, at 84.4% between CORINE and CCT and 86.5% with the MFE. These differences may be due to the fact that most of the area studied corresponds to agricultural land, which results in a much smaller degree of error generated at the boundary delimiting stage by CORINE’s deficient geo-referencing. In those areas where there are major water surfaces the degree of error is extremely high, mainly as a result of the mapping deficiencies in the MFE mentioned above. CORINE only coincides by 25.3% with the CCT and 20.5% with the MFE. This comparison reveals that geo-referencing errors lead to an inadequate comparison of polygons.
35In the comparison of the various percentages in each class studied, the following results have been recorded:
Agricultural surfaces
36The results are shown in Figure 8.
37As it can be seen, herbaceous dry-land crops do not show significant variations, though major changes do exist in irrigated land crops and vineyards. The lower rate in the surface area of irrigated land may have been due to the overexploitation of the aquifers in the West of La Mancha which has led to the decommissioning of irrigated plots and the resulting compensation paid to the farmers affected in accordance with the Wetlands Preservation National Program, approved and started in Spain in the last decade, which fosters the decommissioning of irrigation facilities.
38A further cause might be the inclusion in CORINE of irrigation zones within Class 242 (Complex cultivation patterns). This class is not sufficiently defined and seems to include a variety of land uses which makes its differentiation rather difficult.
39The decrease in the surface area allocated to vineyards is possibly due to the Agrarian Policy of the European Union which has been fostering the uprooting of vineyards in this Region for several years.
Figure 8: Agricultural surfaces: CORINE compared to CCT.
Wooded surfaces
40The comparison between forestry classes shows greater accuracy. In this case, the polygons of the MFE were translated with no serious problems to CORINE categories. The results obtained are shown in Figures 9 and 10.
41In the diagrams we can see, first of all, a decrease in the agricultural surface area as already recorded in the comparison at the second level. The most significant changes are to be found between the classes ‘Transitional woodland shrubbery’, ‘Sclerophyllous vegetation’ and ‘Broad-leaved forestry’. In both databases their total surface area amounts to approximately 12%, with no great differences in the others. We can therefore assume that there is a shift of surfaces from the first two classes to the class ‘Broad-leaved Forestry’. This may be the result of a difference in the criteria adopted when it comes to classifying a formation as a ‘forest’. We can see how this results in the ‘Broad-leaved Forestry’ class increasing from 1.56% to 8.92%.
42A similar shift also seems to exist between the classes ‘Land principally occupied by agriculture, with significant areas of natural vegetation’ and ‘Agro-forestry Areas’, where we can see how both classes make up a total of approximately 4.5%. The difference recorded in the diagrams may be caused by the ambiguity found when it comes to differentiating these two classes.
43On the other hand, the class ‘Mixed Forestry’, which in CORINE has a percentage of only 0.26% and, for this reason, has been included under the heading ‘other’, is given 2.19% by the MFE. Many coniferous formations have bush undergrowth made up of trees belonging to the Quercus Genus and their current status of protected species has led to the growth of mixed forests which, in our area, basically consist of pine (Pinus halepensis Mill., Pinus pinaster Ait.) and evergreen oak (Quercus rotundifolia Lam.) trees. A lower rate can also be observed in the surface area dominated by conifers (pine and Juniperus thurifera L. forestry in the zone studied). This is where these formations may have become part of the ‘Mixed Forestry’ class.
44Another class which has also been included under the heading ‘Other’ is that of ‘Sparsely Vegetated Areas’ which, in our area of study, represents the steppe-like formations made up of gypsophila species. In CORINE there is a lower percentage of the latter probably because these unusual ecosystems are not included in its database.
45Finally, the classes which refer to water surfaces (also included under ‘Other’) show the deficiency mentioned above and already pointed out in the MFE mapping which results in their percentage being lower.
Table des illustrations
Haut de pagePour citer cet article
Référence électronique
Martín de Santa Olalla Mañas, Cabañero Soria et Artigao Ramírez, « Validation of the CORINE land cover database in a pilot zone under semi-arid conditions in La Mancha (Spain) », Cybergeo: European Journal of Geography [En ligne], Cartographie, Imagerie, SIG, document 235, mis en ligne le 07 février 2003, consulté le 08 octobre 2024. URL : http://journals.openedition.org/cybergeo/4233 ; DOI : https://doi.org/10.4000/cybergeo.4233
Haut de pageDroits d’auteur
Le texte seul est utilisable sous licence CC BY 4.0. Les autres éléments (illustrations, fichiers annexes importés) sont « Tous droits réservés », sauf mention contraire.
Haut de page