Monitoring biomass using polarimetric multi-frequency SAR

Classification of radar images remains a challenge due largely to the effects of speckle. In this paper the classification of radar images is accomplished in two steps. First, an image is partioned into uniform areas or segments and second, these segments are then classified into information classes. Both segmentation and classification are achieved by using the Gaussian Markov random field model. The results of segmentation and classification routines applied to the woodland areas in Queensland using both spaceborne and airborne SAR image data are presented. In terms of segmentation, regions whose mean differences are as small as 0.5dB and with ratios of the standard deviation to the mean as high as 0.35 are separated with accuracies approaching 90%. In terms of classification, there are more ambiguities in single-band data. Multi-band polarised data on the other hand provided better results. Relationships established between component biomass and the backscatter coefficient at all wavelengths and polarisations indicated a strong correspondence (r2>0.80) between AIRSAR L- and P-band backscatter and above ground, branch and trunk biomass. The highest correlation between component biomass and backscatter using JERS1 related to the estimation of stem biomass leading to the conclusion that more reliable estimates of total and component biomass will require multiband-multipolarimetric data.