Many studies have used remote sensing to discriminate between mature forest and non-forest areas to determine the extent of deforestation in tropical regions (Tardin et al, 1980; Alves et al, 1992). More attention is now being payed to the remote sensing of secondary forest regrowth because it may have an important part to play in the region's carbon balance (Luckman et al, 1996).
With its all weather capability, microwave sensors guarantee image availability for a region that presents intense cloud cover most of the time, as is the case of most of the Amazon Basin. Another strong motivation for using radar data in the study of second-growth succession is the sensitivity of radar backscatter to vegetation structure which is related to standing biomass (Luckman et al, 1996; Le Toan et al, 1992; Beaudoin et al, 1992).
The objective this paper is to analyse the nature of the relationship between the second-growth succession stages and SIR-C SAR data (Space Shuttle Imaging Radar, L- and C- bands), over the Tapajós region, in the Amazon Basin. The SIR-C SAR data available in this region were bands L and C, HH and HV polarizations. The analysis is accomplished by mapping the regeneration stages using a ten consecutive annual sequence of Landsat TM data (Sant'Anna et al, 1995), and relating these stages with some measures of radar backscatter.
Among the measures that could be derived from SAR images (for instance, see Luckman et al, 1995) two very simple ones are used in this paper: the tonal mean and the coefficient of variation (CV). One reason for using both of these measurements is that in Yanasse et al (1993) was observed that the tonal mean for SAREX (South American Radar EXperiment) data was not sufficient to separate forest from deforested areas, while the CV showed up as a very interesting measure for that purpose. Since SIR-C and SAREX data are quite different it is interesting to analyse whether this result is also valid for the former.
This work uses a regeneration stage map derived from a sequence of Landsat TM images, and the aforementioned measures are calculated for each stage for the whole data set and for individual polygons. These polygons, or image regions, are subjected to an erosion/deletion process in order to avoid spurious observations. Separability among successive regeneration stage classes is assessed with the Bhattacharyya distance for the global means, and the Eucliadean distance for the coefficient of variation.