VEGETATION - 2000
Maggiore - Italy, 3-6 April 2000
A new vegetation
map of Central Africa
Update of the JRC-TREES map of 1992 with SPOT-VEGETATION imagery of 1998
Eerens, Bart Deronde & Jan Van Rensbergen
Centre of Expertise on Remote Sensing and Atmospheric Processes (Vito
Boeretang 200, B-2400 Mol, Belgium.
Tel:(+32) 14 336844 Fax:(+32) 14 322795
Internet: http://www.vito.be E-mail: firstname.lastname@example.org
(pdf file, 842 k)
the frame of a short-term feasibility study (2.5 man-months), the land
cover in the Central-African region was mapped with recent imagery of
the 1km²-resolution sensor SPOT4-VEGETATION. The work was performed on
behalf of METAFRO-InfoSys, the information system of the Belgian Royal
Museum of Central Africa. The image classification was calibrated with
information from the well-known TREES-map, which was established by the
EU-Joint Research Centre on the base of (mainly) NOAA-AVHRR imagery of
1992. In this light, the actual map should not be considered as a new,
stand-alone product, but rather as an update of this TREES-map.
used VGT-imagery comprised the 36 decadal syntheses ("VGT-S10" products)
ranging from April 1998 until March 1999. This yearly image set was pre-processed
in the following way. First, for each decade, the Red and NIR reflectances
were combined into a modified version of the "Soil Adjusted Vegetation
Index" (SAVI), which is less sensitive to variations in the reflectance
of the soil background than the classical NDVI. In order to remove the
cloud perturbations, the SAVI time series were then submitted to a cleaning
procedure and from the smoothed curves, monthly mean SAVI-values were
computed. Finally, these monthly profiles were used to derive "phenological"
images, which quantify the general shape of each pixel's growth curve
by means of parameters such as: the annual SAVI-mean, -extremes and -range,
a seasonality index (mean-weighted amplitude), the start and length of
the growing season, etc.
image classification was realized by means of a Maximum Likelihood algorithm,
applied on a subset of these phenological images and supervised with training
areas selected from the JRC-TREES map of 1992. The resulting land cover
map was embellished with vector information (boundaries, rivers and roads)
and plotted on scale 1:4.000.000. Statistical tables with the acreage
distribution of the land cover classes were also derived for the national
and regional levels, and for both years (1992: TREES vs. 1998: VGT-update).
comparison pointed out that both maps agree fairly well (89% of the concerned
acreage), which implies that no dramatic changes have taken place in the
course of the last six years. However, as the updated map was not checked
on the field, it remains unknown to what extent the observed deviations
(11% of the pixels) are due to misclassifications or to real changes.
Although part of the observed deviations are certainly artefacts, a number
of probably significant land cover changes were revealed which deserve
further inspection, either by field controls or by the analysis of high