Tropical forest cover monitoring: Estimates from the GRFM JERS-1 radar mosaics using wavelet zooming techniques and validation

Abstract

The usefulness of the Global Rain Forest Mapping (GRFM) radar mosaics over South America for tropical forest mapping is assessed in quantitative terms. Estimates of the forest cover are derived from the GRFM mosaic over three test sites that are representative of different fragmentation patterns in Amazonia. Since classical clustering techniques are ill-suited for the GRFM dataset, a novel unsupervised segmentation technique, based on a wavelet frame that acts as a differential operator, is proposed. Validation of the radar-derived estimates is obtained using as reference data thematic maps derived from Landsat Thematic Mapper (TM). The classification accuracy, measured by confusion matrices between the radar and optical derived estimates, are found to be proportional to the landscape spatial fragmentation index. The results from the wavelet classifier are compared with those from classical ISODATA algorithm, and the improvement in accuracy given by the new method is found to be statistically significant. Also the source of omission and commission errors of the radar classification with respect to the reference TM data is discussed. A method based on mapping spatially the confusion between classes is used for the purpose. The study indicates that, within the stated accuracy limit and within the thematic context of tropical forest cover mapping, the GRFM radar mosaics offer with respect to optical data a viable alternative source of information, which however is potentially more powerful for upscaling of the results to continental scale due to the all weather capability of the radar instruments.