In this paper an analysis of the relationship between SIR-C data (acquired in October 1994) and a map of regeneration stage (as obtained from an analysis of ten consecutive annual Landsat TM images up to 1993) was presented. The measures studied were the mean and the CV, observed on a class basis and also for each individual image region after erosion and deletion of small areas.
A quantitative assessment of the separability between consecutive regeneration stage classes was performed using the Bhattacharyya (for the global means) and the Euclidean (for the global CVs) distances.
It is shown that the separabilty among regeneration stages is difficult when individual image regions are used. This discrimination appears to be possible when a large sample is available, and its mean is calculated from the L-band data. In particular, the L-HV data seems to carry more information for this discrimination. The maximum difference of means among classes is of about 5 dB and the means differences for consecutive classes are of about 1 dB for the L-HV data. These conclusions are backed by the Bhattacharyya distance, that suffers reductions of the order of 
when individual observations are used instead of global means.
On the other hand the CV, a measure related to texture, is less suited for this discrimination task. Measures of the CV within the L-band data suggest that some discrimination is possible among early stages of regeneration (less than six years of regeneration). The estimated CV seems to help the discrimination between forest and old regeneration classes.
These results show the importance of having a crosspolarized L band orbital sensor for the discrimination of regeneration stage classes, since the L-HV band and polarization turns out to be the most suited univariate data set for this separation purpose.