Estimation of Coniferous Tree Biomass from High-Resolution Airborne Synthetic Aperture Radar based on the Second-Moment of Cross-Polarization Image Intensity

摘要

The present paper describes a simple technique to measure the coniferous forest biomass by high-resolution synthetic aperture radar (SAR). To date, two approaches to tree biomass estimation by SAR exist. The first method is based on the regression relation between the backscatter radar cross section (RCS) and "ground-truth" biomass, and the second is to utilize the dependence of non-Gaussian texture of SAR images on the tree biomass. The first RCS method is simple and widely used, but there are upper limits of measurable biomass, which depend on the radar frequencies. Lower the frequencies are, higher the biomass saturations become. The texture-based method extends the saturation limits, but requires a probability density function (PDF) which fits best to the image amplitude. The proposed new method is based on the regression analysis between the second moment of image intensity and the measured biomass from the ground survey. To test the theory, we used the L-band cross-polarized SAR data acquired by the airborne Pi-SAR over the coniferous forest in Hokkaido, Japan. A model function is first derived from the regression analysis of 19 forest stands, and the accuracy is estimated by comparing the model-based biomass of 21 other stands of known biomass, yielding the model accuracy of 85%. This accuracy is in the similar order of the previous texture based model. The main advantage of this new method is no requirement of finding a best PDF to fit the image, and can be applied to any images having non-Gaussian texture. In September 2004, the substantial number of trees of the Tomakomai forest fell by a typhoon; the results of the extraction of the typhoon-damaged area are also presented in this paper.