Validating backscatter models for CARABAS SAR images of coniferous forests

Abstract

This paper describes a quantitative comparison of previously developed models for the relationship between CARABAS VHF-band SAR measurements and coniferous forest parameters. On level ground, four models of the backscatter have been studied. For stem volume retrieval, we propose using the normalized backscatter amplitude or the square root of the backscattering coefficient rather than the backscattering coefficient. The normalized backscatter amplitude and the square root of the backscattering coefficient show high correlation with stem volume over the investigated range. In addition, the standard deviation of multiple images of these two measures was found to be independent of the stem volume. For single-image retrievals, the estimated standard deviation in stem volume varies from 32 to 74 m3·ha-1 for forest on level ground, with stem volume ranging from 100 to 600 m3·ha-1. The precision is improved when multiple images with different headings are averaged, since the noise is statistically random between images. On sloping ground, the observed variations in backscatter of images with different headings are larger than those on level ground because the backscatter is affected by the topography. In previous work a model has been derived to describe the effect of a sloping ground plane on backscatter but was not validated due to lack of data. In this paper, 10 CARABAS images acquired from different directions are used to show that the model provides an adequate description of the effects of topography, i.e., the variations in backscatter that remain after using the model are equal to those observed on level ground.