GlobCover 2005 validation dataset
The GlobCover 2005 validation dataset was built relying on a network of international experts, selected for their undisputed expertise on land cover over relative large areas, their familiarity with interpreting remote sensing imagery, their commitment, their complementarities to the other experts and their belonging to well-known international networks.
Experts have used a dedicated working environment for on-screen collection of 'ground truth' data. The validation samples were automatically overlaid either in Google Earth allowing a rapid access to recent remote sensing images with zooming capabilities. In addition, NDVI profiles (8 annual profiles from 2000 to 2007 and the corresponding average profile) were made available for each validation sample, thus complementing the interpretation of the high resolution imagery by its seasonal dynamics.
The GlobCover validation data set contained 4258 samples, globally distributed and selected according to a random stratified sampling. For a given sample, the expert saw not only the sample point but also a box that coincided with the so-called observational unit corresponding to 5x5 MERIS pixels (225 ha). The effective observational unit was not necessarily a square or a circle around the point. Some land cover classes, notably lakes and wetlands, can indeed be rather elongated and but cannot this form should not be discarded because of the shape of the observational unit. The main purpose of the box was to give an idea of the extent of an area of 225 ha.
The experts could describe up to 3 land cover types for each observational unit. A sample was labelled with a single dominant land cover type when more than 75% of the observational unit belongs to the same type. If 2 or 3 land cover types covered each between 25 and 75% of the observational unit, they were described as well. In addition, the experts were invited to give their level of confidence for the labelling.
Sample of consolidated points of the GlobCover 2005 validation dataset
The proposed dataset is the result of a consolidation work realised on the original ESA-GlobCover 2005 dataset. This work was realized as follows. A total of 500 samples were randomly selected and re-interpreted using Google Earth imagery. High confidence was found for almost all samples associated with a single dominant land cover type, which corresponded to 186 points.
The current consolidated dataset should therefore be used with caution. First, it is a subset of the original GlobCover 2005 validation dataset and second, it only contains " pure " samples. The dataset contains a field 'Confidence' describing the level of confidence of the consoliadtion (1:'certain', 2:'reasonable', 3:'uncertain').
This subset should be considered as a simplified dataset. The remaining points will be made available in the near future and at that time only, the consolidated dataset will really become suitable for validation.
The GlobCover 2005 map, legend and description report can be accessed here : http://due.esrin.esa.int/globcover/
References:
Defourny, P., Schouten, L., Bartalev, S., Bontemps, S., Caccetta, P., de Wit, A. J. W., Di Bella, C. M., Gerard, B., Giri, C., Gond, V., Hazeu, G. W., Heiniman, A., Herold, M., Knoops, J., Jaffrain, G., Latifovic, R., Lin, H., Mayaux, P., Mucher, C. A., Nonguierma, A., Stibig, H. J., van Bogaert, E., Vancutsem, C., Bicheron, P., Leroy, M., and Arino, O.: Accuracy assessment of a 300-m global land cover map: the GlobCover experience, in: Proceedings of the 33rd International Symposium on Remote Sensing of Environment, Stresa, Italy, 4-8 May 2009, PS-B1-3(Ref 863), 2009.
Defourny, P.; Mayaux, P.; Herold, M.; Bontemps, S. (2012). Global Land-Cover Map Validation Experiences : Toward the Characterization of Quantitative Uncertainty. In : Remote Sensing of Land Use and Land Cover : Principles and Applications, p. 207-223.
Bicheron P., P. Defourny, C. Brockmann, L. Schouten, C. Vancutsem, M. Huc, S. Bontemps, M. Leroy, F. Achard, M. Herold, F. Ranera, and O. Arino, GlobCover 2005: products description and validation report, ESA GlobCover project, 2008. Available at: http://due.esrin.esa.int/globcover/
Access to the data
We provide an acces to 186 consolidated samples. Each sample plot is described by its ID, the X and Y coordinates of the centroid point (WGS84, in decimal degrees) and the GlobCover class number/name.
You can download this dataset in different formats via the buttons provided at the bottom of this page.
Spatial distribution of the randomly selected sample plots of the GlobCover 2005 dataset available on the the portal
