Pastagens Abandonadas na Amazônia Central Quantificadas com Imagens Landsat

Abstract

Inspection of Landsat images near Manaus in the central Amazon showed a marked difference in the success of pastures between two terra firme areas, one 50 km north of the city on Tertiary deposits in the Suframa Agricultural and Range District (Distrito Agropecuário da Suframa DAS), the other 50 km south of the city on Quaternary deposits, along the road to Autazes. Time series of images, using bands 3, 4, & 5 of the Landsat 5 TM sensor, were used to quantify the distinct success of pastures in these two areas. Three study sites were chosen: the area of influence of the BR-174 Highway in the DAS, the area of influence of branch roads in the DAS, and the area of influence of State Highway AM-254 south of Manaus. Masks were developed to restrict the study to a 4 km wide buffer zone along each road and to eliminate pixels contaminated with cloud or cloud shadow within this zone on any date. Six dates between 1985 and 1999 were studied in the two sites of the DAS, and two dates, 1989 and 1999, for the AM-254 Highway. Pre-processing steps included: geometric registration within each time series; radiometric correction of the DAS series; transformation to top-of-atmosphere reflectance; and median filtering of the transformed bands. Estimation of the areas and rates of deforestation were made by visual interpretation of contrast-enhanced false-color composites. The areas and rates of pasture abandonment (area of regrowth + overgrown pasture + perennial crops), as a percentage of the deforested area, were determined using vegetation indices calculated from bands in reflectance units, with and without radiometric calibration to a clean standard atmosphere. A Maximum Likelihood supervised classifier and a K-Means unsupervised classifier were also employed, always using bands in DN, calibrated to a clean atmosphere in the DAS. North of Manaus, deforestation rate from 1985 to 1999 was 0.34% yr-1 along the BR-174 Highway and 0.19% yr-1 along the branch roads. Along the AM-254 Highway, the rate for 1989-1999 was 0.89% yr-1, about three times higher than the BR-174 Highway within the DAS. By 1999, the total deforestation along branch roads of the DAS was 22% of that study zone; along the BR-174 Highway study zone it was 34% and along the AM-254 Highway it was 37%, suggesting that both ease access and geology/soils have an influence on attactiveness for implanting pastures. The median filter decreased estimates of clear pasture area using vegetation indices, probably by absorbing small clumps of colonizing shrubs into the surrounding pasture matrix. For this reason, all estimates of areas used unfiltered bands. A given vegetation index using a fixed threshold gave different results for the area of clear pasture, using the same bands before and after radiometric calibration to a clean atmosphere. NDVI (normalized difference between TM bands 4 and 3) was most sensitive to this problem; MIR-VI (Middle Infrared Vegetation Index: normalized difference between TM bands 4 and 5) was moderately sensitive; and two SAVI indices were the least sensitive. The vegetation indices which appeared to best separate clean pasture, by thresholds with the least confusion with other cover types, were MIR-VI and SAVI 2 (L=0.75). Rates and areas of pasture abandonment were therefore calculated with these two indices, giving the following results. Pasture abandonment over 14 years in the DAS was higher along the branch roads than along the BR-174. The lowest rates of abandonment were found along the AM-254 Highway. The percentage of deforested area occupied by clear pasture was much higher along the AM-254 Highway, indicating that pastures are much more successful there. By 1999, along this southern highway the percentage of deforested area considered abandoned pasture (regrowth + perennial crops) had attained 52% or 65%, using subjectively chosen thresholds for MIR-VI or SAVI 2, respectively. Along the BR-174 Highway, using the same thresholds and indices, the area abandoned was much higher: 92% or 91%, for MIR-VI or SAVI 2. Percent abandonment was even higher along the branch roads of the DAS: 96% or 94% of the deforested part. The percentage of the entire study zone occupied by clear pasture in 1999, using MIR-VI , was 18% along the AM-254, but much lower in the DAS: 2.6% along the BR-174 Highway and only 0.8% along the branch roads. These numbers again demonstrate the much higher success of pastures to the south of Manaus and the failure of the DAS for this type of land use. The supervised classifier used a single set of six training areas and radiometrically intercalibrated bands over the six date series of the DAS, but used two separate sets of training areas in the two uncalibrated images of the AM-254 Highway. The classifier showed a tendency of pasture abandonment in all three study areas. The unsupervised classifier did not give reliable results, as it was restricted to only six classes. Some of the more spatially extensive land cover types, such as primary forest, captured two or three of these classes leaving an insufficient number for the other cover types of interest, such as clear pasture, overgrown pasture and secondary forest.