This paper aims to develop a methodology of multisensor integration for an Amazon monitoring system. The proposed system employs the Vegetation Continuous Fields (VCF) method that uses the decision tree algorithm. The algorithm uses a set of independent variables, in this case MODIS multi-temporal metrics, to recursively split a dependent variable, in this case training data from class memberships, into subsets, which maximize the reduction of squares of sum of the residuals. The training data are obtained by high-resolution imagery classification (Landsat/TM, ETM+ and CBERS 2/CCD). In this study, an automated algorithm was developed from IDL language in the ENVI software and the statistical procedure of the S-PLUS software. The study area is Mato Grosso State with an extensive area of Amazon forest. The scenes are classified in three classes: forest, non-forest, and water. Comparisons of the final product with regional land cover maps derived from PRODES revel general agreement. Therefore, the results of this study suggest that the methodology is appropriate for land cover determination in the Amazon forest.
remote sensing; digital image processing; continuous fields; regression trees; Amazon forest
