Chris Bellman
vector for the class is updated. This second step is repeated until the list does not contain any other region eligible for
class detection. In the third step, the regions are classified by using the mean vector of each class as the initial centers
for the k-means algorithm. Chi-square acceptance threshold of 99%, 95%, and 90% were tested. Based on visual
analyzes of the classified images on the screen, a 90% acceptance threshold was selected as the most appropriate to
define eroded areas. Due to the low spectral contrast between eroded areas and natural open grass fields in the
Tepequem plateau, ground truth information was very important to define classes in the region-classified images. In this
process some classes were regrouped to express more faithfully the terrain features.
4. RESULTS AND DISCUSSIONS
Figure 2 shows a color composite and the Landsat-TM time series covering the region of the Tepequem plateau. Hues
of magenta and green in the color composites indicate, respectively, areas of open grass fields over the plateau and areas
of tropical rain forest in the surrounding. Exception for a few clouds, bright areas in the Landsat-TM band 3 images
indicate degraded terrain associated with gold mining activities (images cover 15 x 10 km).
Figure 2. Color composite and Landsat-TM time series of the study area. Hues of magenta and green in the color
composite indicate, respectively, areas of open grass fields over the Tepequem plateau, and areas of tropical rain forest
in the surrounding; bright areas in the Landsat-TM band 3 images indicate degraded terrain associated with gold mining
activities (images cover 15 x 10 km).
Thematic maps showing distribution of degraded areas are showed in Figure 3. Up to 1987, gold mining areas
predominated along the stream that drains the northern part of the Tepequem plateau. Small mined areas also occur
downstream outside of the plateau, in areas of tropical rain forest. Data from 1991 indicated that mining activities had
been extended to streams in both central and south parts of the Tepequem plateau. Such activities continued growing in
the region, according to the data from 1994. Based on the maps, degraded areas increased by 388% from 1987 to 1994,
comprising 1,070 ha, 2,550 ha, and 4,150 ha, in 1987, 1991, and 1994 Landsat-TM images, respectively.
Due to a governmental action, combined with the exhaustion of the mineralized alluvial deposits, garimpeiros
abandoned the region around 1993/1994. As a result, vegetation soon started recovering eroded areas. Maps based on
1996 and 1999 TM images showed that degraded areas decreased to 3,660 ha and to 3,391 ha, respectively (Figure 3).
Figure 4 shows in green and blue areas of vegetation regrowth, relative to the 1994 map. These areas, situated in the
borders of the gold mined areas, comprised 490 ha in 1996, increasing to 759 ha in 1999. These figures suggest a
recovering rate of 245 ha/year from 1994 to 1996, decreasing to approximately 90 ha/year in the 1996-1999 period. If a
regrowth rate similar to the last remained stable over time, damaged areas would be entirely recovered only in the year
of 2037 (Figure 5). However, recovering process would probably take much more time, because this rate deals with
regeneration occurring mainly in the borders of the gold mined areas (Figure 4), where recovering process may be faster
compared to the core of the mined areas where damage was more intense.
68 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
"P P L
]]