VEGETATION - 2000

Lake Maggiore - Italy, 3-6 April 2000

Sub-pixel characterization of land cover at the global scale
using SPOT-VEGETATION imagery

Else Swinnen (*), Frank Canters(**) & Herman Eerens (*)
*Centre of Expertise on Remote Sensing and Atmospheric Processes (Vito-TAP)
Boeretang 200, B-2400 Mol, Belgium.
Tel: (+32) 14 336844 Fax: (+32) 14 322795 E-mail: jan.vanrensbergen@vito.be
**Centre for Cartography and GIS - Brussels Free University (VUB)
Pleinlaan 2, B-1050 Brussel
Tel: (+32) 2 6293556 Fax: (+32) 2 6293378 E-mail: fcanters@vub.ac.be

Accurate information about the spatial distribution of different land cover types is of utmost importance for the modelling of ecological and environmental processes at the regional and global scale. The objectives of the project, that will be presented in this poster, consist in the development of an easily reproducible method to derive global land cover maps from 1km VEGETATION imagery (full year sets of 10-daily syntheses), and in the inference of accurate estimates of the areal proportion of the different cover types at the level of typical cell sizes used in continental and global scale modelling. The general idea is to develop a supervised method for sub-pixel classification of major land cover types, using high-resolution data for the training. By defining calibration models, relating class proportions obtained from the classification to class proportions obtained from the high-resolution reference data, local estimates of the proportion of different land cover types will be derived for different levels of aggregation.

At present various strategies for sub-pixel classification are tested on a small section of the global VEGETATION data set of 1998-1999, covering most of the European continent. The definition of land cover types is based on the IGBP Global Land Cover classification scheme. For the training, use is made of the ETC/LC production database of CORINE Land Cover, containing data collected between 1989 and 1997. All co-ordinate information in the CORINE database was geometrically transformed to the same reference system as the VEGETATION imagery. Extra adjustments were made to achieve an optimal fit. After spatial aggregation of the CORINE data (44 land cover classes) to the agreed set of major land cover types, the transformed version of the database was gridded at a resolution of 20m. From this detailed information, land cover proportions were calculated for each 1km² pixel in the VEGETATION imagery. This proportional data forms the input for the training and verification of the classification tests that are currently performed. Once a suitable classification strategy has been defined, the same high-resolution data will be used to develop an appropriate calibration method to improve the quality of land cover area estimates derived from the classification.